Will be good to have a significant jump in performance + jump in battery life. Also, wasn't it data center first with new nodes - IceLake Xeons followed by the mobile and desktop or that's not for this gen products?
Yup. And imagine they hadn’t had this big 10nm Delay - a 18% IPC improvement would have made ZEN2 a whole lot less interesting. Alas, it will be another year before these reach desktops and by that time, amd has ZEN2+ out ;-)
One thing that Intel is doing smart is learning from the 10nm delay, they separate the process from the architexture. In the market today desktop is minor market, just a bunch of gaming kids. Mobile is the primary market now
We are consistently refreshing all these, morning, night and night results on our site. You don't have to meander around for getting the Sambad lottery result today, you are most welcome here to check your lottery result on our site effectively. https://todaysambadlottery.in/
That's exactly what they haven't done. They've continued to push 14nm revisions of Skylake over and over instead of backporting Ice Lake / Sunny Cove to 14nm. That's why Ice Lake was delayed so much, it had to wait for 10nm.
So they SAY. Intel says a lot of things, most of them contradictory (right now they are telling us BOTH that there will be speed optimizations on 10nm, AND that there will be a rapid transition to 7nm ?!?)
I’ll believe this story when I see it. Given how slowly Intel changes, even if they did actually agree to this, it may take 5 years before it’s implemented.
Yup, as with anything, it is nothing until its released and retail units (not engineering samples) are independently tested. Then an only then is it real.
It looks like Sunny Cove arch is going to nearly match desktop performance, at least in short bursts, in spite of the lower clocks.
I don't think that's going to hold true on extended runs. Encode a short video and Ice Lake will probably kick ass. Encode a longer video and this likely won't be the case, it'll hit that 25w TDP and drop off.
Their bread and butter is the giant-scale cloud and enterprise market. ironically, they are among the most sensitive to overall power consumption due to scale. Intel began the effort to switch to developing more power efficient cores more than 10 years ago however. They're not the leader because ARM already was, and started from the other direction. You may also well notice that neither has done particularly well moving into each others architectural strongholds.
People need to pay attention to that IPC number. That's only in comparison to older Intel processors, not current gen. They got a few people here clearly with that trick.
Exactly, IPC at the moment is based off skylake CPU cores. All we have got is more cores bursting to high real TDPs. Remember that Intel basically skipped/ditched their traditional “Tick” ; first gen 10nm parts.
This +18% IPC doesn't include the security mitigations, and note the clock speed. From Intel, 15W TDP is for the base speed, which is quite a bit lower than the 14nm chips.
This 10nm delay is pure incompetence. Would this have happened under Moore or Grove? Intel, instead of innovating, keeps on making minor improvements on old designs. They really need a new management team with some imagination.
Rickjason part of the lack of innovation.. could be due to the lack of competition, just like back before the athlon 64 came out, and caught intel with its pants down.. just like with Zen :-)
Usually Intel releases new technology in Xeon's first for example AVX 512 and since most of industry is going mobile now - then it logic to do mobile first. I would expect the Ice Lake based Xeons to have new technology components in it.
most of the industry is goimh mobile ?? yea sure.. prove it... no one i know is buy laptops... they are still looking at desktops.. and for the moment.. they are looking at zen 2.. not intel...
Its true of the general public too. Walk into a retail store that has a history of selling computers and you'll find lots of laptops on display with maybe one desktop PC available but not even out on a shelf. Desktops are still around mostly as gaming systems because of upgrading and bling. Apart from that there are still heavy lifting workstations and then mini PCs (made largely of laptop-class components). Desktops are few and far between which is why there is now a large price premium for components in order to keep margins a shrinking segment high enough to continue pulling profits as sales volumes taper off.
thats not the case here.. the stores may be like that.. but no one i know is interested in laptops... i havent heard any of their friends say they want a laptop.. i even keep getting asked about desktops, and what they should get .... maybe its different market needs for the areas ?
Reality is almost twice as many laptops & 2-in-1s are sold these days as desktops, and the ratio is going to keep going farther and farther that way. You and your friends probably have similar work or interests (PC gaming is the most likely, since you're on a hardware enthusiast site) that makes desktops more useful, but the people you know are not a representative sample of the broader market.
drothgery nope.. various usage.. games, general web browsing, music listening... they just dont have any interest in laptops for some reason... i am yes. but them.. they dont pay attention to hardware news... they are not.. but still.. a few here seem to think every one wants laptops now... which im trying to say... isnt the case ...
No not really. Though laptops provide the convenience, look at any office. Most are connected through desktops. Desktops provide the convenience of easy maintenance, lower costs and upgradability while a laptop is mostly obsolete by 5 years and cannot be upgraded. Most developers, graphic designers and marketing mostly use a desktop. The only profession where a laptop is popular is among students and journalists.
This is not true for all companys I worked for. I work in pharmaceutical R&D and in chemical R&D before thet and litterally every (non production) employee gets a laptop. From assistants over researchers to the CEO. And then some of the scientists have a desktop/workstation for CAD or other demanding work. In addition to their laptop.
Every office I've looked at replaces their machines within 5 years anyway, they don't bother with upgrading desktops. I'm a software developer and have usually worked on a desktop, but on my last hardware refresh I was moved to a laptop.
According the roadmap of intel(include some information by Huawei leaking ),intel will unveils cooper lake-sp in 2019q4 and unveils ice lake-sp in 2020q1 for datacenter. Those production will sell in 2020Q1.
I think it's mostly been desktop, laptop, server. Broadwell flipped the script, pretty certain Skylake was desktop first, Haswell too. Sandy Bridge was Jan 2011, mobiles didn't come out until Sept 2011. Ivy Bridge was released in April 2012, both laptop and desktop.
Based on their own graphs only the 64-core iGPU will provide a significant jump in GPU performance. Due to the low clocks the CPU cores will struggle to provide more than 2% extra performance (over Whiskey Lake-U) at best. To be clear the +18% IPC is over Skylake CPUs, which were released in 2015, not over the current Whiskey/Amber Lake U/Y CPUs.
Oh, there will also be quite a jump in encoding performance, for those who prefer to encode video in hardware (for me that's very important). Decoding videos should also become even smoother, lowering CPU utilization further (no word about AV1 decoding yet though). I don't care about the faster wireless because I don't throw LAN parties, while the benefit of the AI speedup is unclear.
its funnier how you are still pushing intel when its pretty clear that intel.. may NOT be the performance leader anymore... and baseing things on mobile.. is still valid as it could translate to desktop performace still... but we will have to wait to see to be sure
Half the market is still desktops and they still maintain a slight lead over laptops. Also desktops are on the rise bc performance is finally on the rise.
Whiskey/Amber/Coffee/Kaby/whatever Lakes are all Skylake, so identical IPC (excluding security fixes in hardware, but that should really be treated separately from the Skylake architecture's IPC).
Where did you find graphs that compare anything other than the 64-core version? Which BTW "based on their own graphs" seems to be about 60-70% faster than Whiskey Lake-U.
(a) When full Spectre mitigation costs around 15%, that’s not a trivial point in terms of cross-device comparisons..
(b) The IPC comparison is interesting in terms of suggesting Intel’s future, but if you are buying a chip, what matters is performance, not how it’s split between IPC and frequency. AVX512 may matter for future apps, but it’s not unreasonable to point out that these are NOT really faster chips (for most single threaded performance) no matter what Intel says.
The Spectre impact depends on the benchmark, certainly there is a 15% worst case but over a big range of benches the diff is on average more like 3-5%. Certainly on spec which is probably the benchmark intel got the 18% from.
"only the 64-core iGPU will provide a significant jump in GPU performance" " To be clear the +18% IPC is over Skylake CPUs, which were released in 2015, not over the current Whiskey/Amber Lake U/Y CPUs."
Yeah, there doesn't seem to be enough understanding of this. Intel's claiming perf improvements over 4-generation-old processors is a bit disingenuous.
Also, for those relishing the idea that the Intel iGPUs may now be better than the Ryzen ones, remember that AMD can probably do the same thing and add more cores to THEIR APUs as well. (Whether they will or not is another matter, of course.)
Intel already has i7-8569U [4.70 GHz boost] with Iris Plus Graphics 655[921.6 GFLOPS ] with same 28W TDP, so how much better the 10nm part looks like now[at least from 14nm++ down to 10nm]?
In terms of absolute top compute power, probably no difference, neither for CPU nor GPU.
It's all about the Watts and the ability to sustain that max power closer to 15 Watts then 28 Watts, which makes all the difference in the world running on battery and for noise/form-factor.
And btw. good luck finding the current Iris Plus parts in anything but * insanely priced Macs * astonishingly priced Intel NUCs
Astonishing, because Intel doesn't *charge* extra for the much bigger GPU and the eDRAM, they just won't *sell* it to anyone but Apple and themselves.
Somehow Intel must have decided 1. We don't charge *ever* for iGPUs (actual price difference seems around $5) 2. We don't make enough profit selling 48 or 72 EU parts to the open market under rule 1, so we won't
I think you are comparing 8th gen U vs Ice Lake Y - it appears on spec that the new XPS 13 2in1 is Y based processor and not the Ice Lake U processor. I have original XPS 13 2in1 and low based ghz and max ghz sounds like this is Y processor and not U processor.
My thought is that it could be Y as existing XPS 13 2in1, just Intel has increase the core count. the specs indicated 9W, 15W and 28W but only 4 cpu's listed. This likely means more processors are to come and not a guarantee that that one in XPS 13 2in1 is U processor. Also 28W version should be a lot faster because of higher watts.
It hard to tell now what the exact performance will be - but Windows Central states that new Dell XPS 13 2in1 could be best convertible ultrabook of 2019. Notebook check did indicates that it is launch on July 10 - you know about 40 days from now.
One thing that is confused only 4 models list here, 11 models coming and 3 different watts This is summer 2019 release and what is being release for holidays. Intel is probably waiting to see how AMD responds and come back hitting.
Well intel can wait a long time, AMD lacks a compelling solution that competes with this in the (ultra) portable market... it will take a while until zen2 and Navi come to mobile, long enough perhaps that they even end up competing with whatever intel does next. AMD clearly focuses on enthusiast desktop & server for now, where they can win easily. Sadly for them laptops are a huge market... though of course there is more $$$ in servers.
Take a look at the clock speeds on these chips. Yes, turbo goes higher, but at 15W, that is base clocks, and turbo we don't know how much power or heat gets generated. AMD SHOULD move up the next generation laptop chips by a few months if possible, because a 7nm Zen2 based laptop chip would destroy these new Intel 10nm laptop chips.
there is no such thing as an AMD NUC .. (mini-pc) believ me I've looked . A whole segment of the market going begging Under 15w from AMD Mini-Pc would be great ... maybe some day .. when their chip supply improves?
Wasn't the issue with AMD chips something along the lines of not properly ramping up or down, wasting a lot of battery? Once fixed playing field is even at worst, while AMD is releasing an improved APU based on Navi, just no die shrink.
The CPU clocks are going to be quite lower than the 14nm++ equivalents, due to persisting thermal issues and low yields. I have no idea what that means exactly for the iGPU clocks, but I would expect quite lower iGPU clocks as well, or at least much shorter sustained iGPU boost clocks. Gen11's design is somewhat different though, so we will need to wait for devices to be released to find out how the iGPUs will perform (we already know, based on Intel's own graphs, that Whiskey Lake-U is going to have crappy performance - at best 2% faster than Whiskey Lake-U . Their thermal issues also had to force them to blow up the TDP of the Y parts from 4.5W to 9 - 12 W, which is staggeringly high and probably precludes fanless designs, i.e the designs the Y CPUs are actually meant for...).
The TDP growth from 5->9W for the Y parts can mostly be explained by integrating wifi and thunderbolt 3, which are worth 2W each at least. The cores themselves are most likely being allocated slightly less power than before, and system power consumption will improve.
A decent improvement on the graphics front however just barely beating AMD's 3700U isn't that impressive. I mean we are talking a die shrink so there is a lot more die space to play with and graphics should scale out nicely with more graphics cores. AMD's first 7nm mobile chip to replace the 3700U should surpass this 10nm chip rather quickly.
I wouldn't say that AMD will easily surpass this. The jump to 7nm for AMD's APUs will be an interesting one. It appears that AMD is currently content to have their mobile chips be based on mature nodes and aimed at a value position in the market. With that in mind, I suspect that Renoir, which is the code name for the next gen APU from AMD, will be based on the current 7nm process, will likely continue to be a monolithic die, will likely retain the 4 core, single CCX arrangement that they currently use, will likely have 1/2 the L3 of their existing 7nm CCX design (so, 8MB), and will likely have a NAVI based iGPU section, probably with 20 CUs (more CUs running at lower speeds can be more power efficient). The big gain for AMD will be that NAVI is expected to be more efficient with memory bandwidth (new caching structure, supports advanced features that save on bandwidth usage) and have support for faster DDR4 (though, I don't imagine that we'll be seeing anything much faster than DDR4-3200 in an SO-DIMM in wide usage by next year). I don't expect that AMD will have a memory controller that can support 4 X 32 bit LPDDR4X-3733, so will be at a bandwidth disadvantage.
The one possibility that AMD has to surpass Intel here would be for them to leverage their experience with HBM and have Renoir be a design that is capable of being either independently packaged for lower cost applications, or being mounted on an Emib/MCM package and support a "back channel" connection to a 4GB HBM stack. A 20CU Navi with 4GB of dedicated HBM with appropriate clocks would likely be firmly in GT 1650/1660 territory, which would be just fine for 1080p gaming. That seems like a reasonable sweet spot for mobile devices and something that could also see reuse as a desktop APU for Home Theater machines that live in mini-ITX cases.
As Intel's upcoming solution is barely beating existing 3700U devices (in Intel tests), AMD will surely regain lead with 4700U even if it remains on the 12nm process. If it doesn't actually lead already in independent tests. Switching to 7nm is fairly likely next year (keeping just 2C/3CU cheap parts on 12nm) and will greatly surpass IceLake. But by then Intel will finally manage their dGPU. I wonder how this one ends up. So far, GPU engineering hasn't shown anything wow, but with their new push in marketing they might launch something interesting. Or at least their PowerPoint will say so.
Navi + 7nm will likely smoke these Ice Lake parts. Someone has done some calculations comparing Ice Lake to Zen 2 in Geekbench and the IPC advantage is more like 6%, not the 18% that Intel was touting. They must have included AVX-512 (not just integer like AMD did) in the mix to give the appearance of extremely good IPC.
Intel reported in a graph that Ice Lake-U has a 4% higher single thread performance than Whiskey Lake-U, with both at 15W. They did not compare them directly, because apparently it wouldn't look good. They set Broadwell at 1.00 performance, reporting ~1.43 for Whiskey Lake-U and ~1.47 for Ice Lake-U.
What do you think the real number is? I was thinking ~2%, but if they employed AVX-512 to skew the score the actual gains might be in the 0% to 1% range. Basically the question is if their 10nm+ node's very low clocks ate *all* their IPC gains or if it left a few scraps of it. Which is why their main presentation had big graphs shouting about x2 iGPU performance(!), x2.5 AI performance(!) and ... 3 times higher wi-fi "performance"(!), with no word about CPU performance. I smell a fiasco slowly brewing...
That +18% IPC was over the (original) Skylake that was released way back in 2015, not over Intel's current CPUs (i.e. since we talking about a TDP of 15W over Whiskey Lake-U). These are technically still based on the Skylake arch but have been optimized and refined multiple times, are fabbed on newer variants of 14nm etc etc Intel did not report the IPC gain over Whiskey Lake because it is obviously very low, and surely single digit. It is difficult to asses it because Intel did not disclose base clocks, but it should be in the 6 to 8% range.
Whiskey Lake has identical IPC to Skylake because it *is* Skylake. It has some security fixes, but other than that, they are the same. Faster RAM might give the illusion of slightly higher "IPC", but the only real improvement is the improved manufacturing leading to higher clocks. Actually the clocks have been improved so much that since Skylake we have pretty much had the biggest generation-over-generation performance improvements since Sandy. At least running stock clocks.
ajp_anton.. tell that to HStewart.. he has been saying for a while now that what intel is realeasing now.. is a new architeCture.. when all it is.. is yet another update/refresh...
Look at architexture on chip Korguz, and tell me that it not different than Skylake, it has more ALUs and other logic, larger cache and besides that also AVX-512 and other enhancements. It would be pure stupidly to say it is update/refresh on skylake. Of course you maybe talking about WhiskeyLake and not Ice Lake which I referring to.
One difference that I just realize was started on my Dell XPS 15 2in1 Kaby Lake G and available in these new processors in Dynamic Tuning, plus there are other aspects but core IPC performance of Skylake series is same - but they adjusted parts of it.
But I think you were confused between IceLake and other processors - yes we all don't live in desktop line and who knows why Intel calls new desktops 10 series on 14nm.
HStewart.. 1st.. do you NOT know how to spell ??? its architeCture... NOT architexture.. for some one who says he has been using computers for as long as you " say " you have.. you would think you would know how to spell it CORRECTLY.
2nd.. it ISNT'T different then skylake.. its just a refresh/update to an EXISTING architeCture that intel has been feeding us for 6 or 7 YEARS now " it has more ALUs and other logic, larger cache and besides that also AVX-512 and other enhancements. " what do you think and update/refresh is??? Amd going from bulldozer to zen THAT is a new architeCture... what intel is doing is NOT but you are too much of an intel fanboy to realize this. you keep believing intels bs and lies.. wake up HStewart.. stop being the fanboy you are.. and open your eyes
Amused at your response to HStewart for misspelling a word, then you proceed to follow with two misspelled words of your own (or one with another wrong word). Spelled correctly, the second paragraph would start (capitalizing words that are corrected) ...it ISN'T different THAN skylake...
catavalon21 the joy of typing on my phone.. vs my keyboard.. but if you look at ALL of HStewarts posts on here.. he ALWAYS spells architeCture wrong... ALWAYS, thats the difference... im sure you have spelled words wrong ( like any one else ) when typing on a phone.. vs keyboard :-)
Sure, I also make tons of typing errors on a full sized keyboard. I type poorly and often do a crummy job of proofreading. I also know that for many contributors on AT, English is not their first language, and try to keep that in mind when the tendency to get picky hits me.
All the Lakes are the same basic architecture with refinements and enhancements and the of new feature. It's pretty much advertised in the name.
The *wells were based on the same architecture, which changed going into the *Lakes.
No different than how AMD iterated over the Bulldozer architecture all the way through Excavator/Kaveri, then switched to Zen. Then they iterated over Zen with Zen+. Now they have a new architecture with Zen 2.
Maybe next year Intel will have a new architecture to base their chips on that's as different to Skylake as Skylake was to Broadwell. That's when the Faveros and EMIB stuff should get interesting.
If you think back to "the old days", an existing architecture was used across two nodes, then a new architecture was released on the new node (tick-tock setup).
Skylake was the new architecture on an existing node. All the Lakes have been iterations over slightly tweaked optimisations of that node. And nope they finally get the existing architecture running on a new node. Tick-tick-tick-tick-tick-tock.
Next year will be the new architecture (tick) on 10 nm. There may be more ticks on 10 nm while they prepare 7 nm. Then the tock will happen on 7 nm. And the prices will repeat.
Yes, Ivy was a refinement of Sandy, etc. But all of the lakes after Skylake are not even refinements. They have fixes for security and maybe other bugs, but they are all Skylake. Intel changed the whole naming scheme because of this - one name for the chip, one name for the CPU core architecture (Ice Lake now has Sunny Cove cores).
Kaby Lake, Coffee Lake, Amber Lake, Whiskey Lake all have Skylake cores. No tweaks, no real improvements, only bugfixed Skylake cores with identical performance. The only exception to this is Cannon Lake, which actually did have a new core (though it's the same kind of tweak that Ivy and Broadwell were).
HStewart, IF this really was a new architecture, then WHY is intel THEMSELVES calling this 10th gen ?? if it was a new architecture like you wish it was... then wouldn't it be called " intel gen 1 10nm ( insert new code name here ) ??? this is just an update/refresh that intel has been using as the base for their cpus for the last few years as korguz has said. Zen 2, is just an update/refresh of Zen+, which is an update/refresh of the base architecture that was Zen, plan and simple... lets see you try to spin this to make your intel look better then it is.
I never mentioned Ice Lake, why did you think I did?
Ice Lake is finally new. Not new as in being a departure of the "Core" architecture, but it's finally something new.
Skylake, Kaby Lake, Coffee Lake, Amber Lake and Whiskey Lake all use the same Skylake core with identical IPC. Only bugs have been fixed, which may have impacted performance *later, if they are being fixed in software*, but the base IPC is identical. Cannon Lake was the tweak to Skylake that previous tweaks have been (Ivy Bridge, Broadwell), but it was never really released. Ice Lake (using Sunny Cove cores) is supposed to be a new architecture (still within Core), as Haswell and Skylake were.
When you figure that compared to Sky Lake the Zen+ was 3-5% lower in IPC that would put Zen 2 at 10-12% higher IPC than Sky Lake, if the quoted 15% IPC increase is true. Therefore if Ice Lake has an 18% IPC increase it will be about 6% higher IPC compared to Zen 2. None of this will be known until systems are out. Also Zen 2 won't be doing battle with Intel for a while since Ice Lake is only in laptops and Zen 2 is only desktop.
Well.. Amd is getting some huge play from the OEMs for gaming laptops over the next year.. so you can bet they have a few innovations which are pulling those OEMs over. Supposedly going to see some fairly high end laptops with Amd inside.
Still any improvement on graphics by intel is always welcome news.
I just hope all of these systems with ONLY a 128GB SSD for storage go away, because your typical consumer has no clue, and then freaks out when their storage fills up within a month of purchase due to pictures, music, or whatever.
Desktop chips can be installed on release day, so that means you will see consumers with the new generation on day one, talking about the performance improvements. Laptop chips will take a minimum of three months to go from sending them to OEM laptop makers to QA, BIOS adjustments, cooling solutions, etc. Then another three months before the shipping products actually get into the hands of consumers. Even if OEMs start shipping laptops on launch day, it will still take three months before consumers will be able to get them.
Since desktop chips can also go into existing motherboards, you will see a lot of upgrades as well from first to third generation. The marketing splash of enthusiasts talking about the new chips does help drive sales in laptops, but there will always be a bit of a delay.
You also have the effect of Navi...the next generation laptop chips from AMD will probably have Navi instead of Vega graphics, so working the kinks out of the drivers for an extra 3-5 months will make for a better end-user experience.
With the 7nm process for higher clock speeds, the +13-15% IPC, and Windows support being more mature, I'd say that AMD has a good chance to beat Intel in laptop performance as well. Intel 10nm production is going to remain low, while TSMC 7nm is going to be able to churn out chips for AMD in much higher numbers.
Took Intel long enough, but now Qualcomm Snapdragon 8cx is coming with a 5G modem for always-on computing. Intel will be tough to beat that unless their Ice Lake chips are cheaper than Qualcomm's.
What is the relevance to speak about a Qualcomm SOC unable to run native x86 applications? Apples and oranges. Snapdragon 8cx is only an exercise of a secondary Qualcomm team with a bad future ahead
It can run x86. It's emulation, so a bit slower, but it can. And the performance is impressive considering it is emulation. That's how it starts, with lower performance but "everything works". Once (or rather if...) it gets momentum, more applications might get native Arm64 versions.
Not quite the same market, but yes I agree and I'd love to get one of those 8cx Lenovos into my hands!
Not so sure if I'd run Windows on it though, especially not if it's one of those M$-Store locked-in variants. Linux or Android desktops seem much more attractive and compatible with what I'd use it for.
Dual-boot my Whiskey Lake with WinX and Fedora/Cinnamon.
Windows on ARM is not emulated, it emulates X86 written applications. Any native ARM applications will work really good. just not many ARM based apps yet for Windows. They're probably coming though.
Looks like the 3rd article about Ice Lake and comparing it Ryzen 3700u and Vega 10. And still no mention in the article about how Intel used more expensive RAM to get 56% more bandwidth and on par performance. Ryzen 3700u using RAM would get 33% more performance and walk all over the Ice Lake iGPU.
That excuses to some degree Intel on a technicality but certainly not reviewers. Reviewer should know better than to just take these great leaps of performance on face value from Intel without mentioning the discrepancy comes from RAM speed, not the improvements in architecture. These sites should be doing tech journalism not masked payed advertising.
Well, since you can't really control the RAM speed in mobile devices, the source of the performance might be interesting, but for and end user also, ultimately, irrelevant, no?
There is no need for "excuses". It is AMDs technical limitation in the very same way, that intel is for example limited by compute units count in current generation of iGPUs compared to AMD. You can say "if intel had same ammount of compute units as amd, they would be faster and it is unfair that amd has more compute units!" but you can only blame intel that they did not put more of them. Same goes for lack of RAM support on AMD part.
Ian seems busy, as it seems he didn't do any own analysing. He only repeated what was shown in the slides. I'm looking forward to the deep-dive he promised.
"If looking at the geometric mean for the tests run today, the Intel systems all saw about 16% lower performance out-of-the-box now with these default mitigations and obviously even lower if disabling Hyper Threading for maximum security. The two AMD systems tested saw a 3% performance hit with the default mitigations. While there are minor differences between the systems to consider, the mitigation impact is enough to draw the Core i7 8700K much closer to the Ryzen 7 2700X and the Core i9 7980XE to the Threadripper 2990WX."
Read his comment carefully. He's saying due to the Hardware patches, they were able to regain 15% IPC lost to the mitigation. Keep in mind Intel is comparing vs Skylake, not Whiskey Lake, which if Intel is to be believed, Whiskey Lake had some hardware fixes for Meltdown and other vulnerabilities.
So his comment isn't BS. It's a play on the fact the software/microcode fixes caused slowdowns. And because of the hardware fixes in Ice Lake, it regains that, plus 3%.
The problem is that we don't know if Intel is comparing old Skylake performance pre-mitigations. Intel did say that not all mitigations were in place when that +18% IPC figure was done. Software/microcode fixes will still result in performance degradation, so honestly, at this point we can't really know how much of an improvement Intel has come up with. We DO know that since these are laptop chips, no consumers or review sites will be able to test them for at least three more months.
The IPC increase is due to the cache increases and it appears other improvements to the back end. It's similar to how Apple has been increasing performance for its in house chips over the past several years. Give it more cache, make it wider, watch it considerably outperform its predecessor.
The IPC improvements are visible over Whiskey Lake which does have hardware mitigations. The geekbench scores show upwards of 20% IPC gains for single core but the frequencies of Ice Lake CPUs are a lot lower and thus the performance is roughly the same as Whiskey Lake.
Perhaps not exciting for everyone, but I'm hopeful because of the TB3 integration. I'm curious to see how this impacts the design of the Surface Pro 7. Unfortunately we're probably waiting until Spring 2020 to find out.
18% IPC increase but vs 15W i7-8665U the turbo frequency is down 900 MHz and base frequency is down 600 MHz (in case of the 1065 shown in the table). Going by turbo frequencies: 3.9/4.8=0.8125 0.8125*1.18=0.95875
So it is slower than i7-8665U. Only at the maximum 4.1 GHz, will speed parity be achieved. Although, yes, I am comparing it to the processor with the highest frequencies and hence price.
Even the 8th Gen i5's have higher boost than 4.1GHz. Across the board the 10th Gen has lower clocks than the 8th Gen. Sure IPC increases will help mitigate the lower clock speed, but it is alarming how much lower the clocks are on 10nm+ from Intel. These low clocks are probably why there isn't a desktop CPU being released since it won't clock high enough to overtake the 9th Gen in absolute performance.
Saw an interesting lecture the other day. I didn't realize that all things being equal (ie you don't introduce a new structure like FinFET) that transistor performance no longer scales with nodes. You go to a smaller node you get a slower transistor, since the voltage can no longer scale at the same rate. So maybe something brilliant comes along but it looks like peak clock frequencies will be coming down. The only reason to scale at this point is economics, and even that is starting to become questionable if the large investment now required for the next node cannot show returns.
Exactly, and it's been like that for many years already.
Actually the 4.8 GHz Turbos on current Whiskey Lakes are just downright crazy and the result of Intel engineers spending years on getting that type of performance via physical tuning out of an "old engine". Those speeds also require 25 Watts on my i7-8565U, so they aren't sustained beyond a couple of seconds.
Most of that physical tuning simply cannot be carried over to a new process node: With EUV and everything else, it's just a completely new ball game.
Density gains hopefully will pay off the additional fab cost and allow widening those CPU structures for the IPC gains, additional cores and beefier GPUs.
Speed and density used to be linked in the good old days up to perhaps 45nm, but they have become rather antagonistic.
I won't predict speeds won't climb to 5GHz again (evidently you just need to leave the engineers alone with a new process node for long enough), but it's going to be harder with every shrink.
Not just performance, but cost/transistor has been increasing per node since 28nm. Expecting die shrinks to give you cheaper faster chips has not been true for many years now.
No it has not! If that were true, what do you think would be the cost of, say, an A12X chip with 12 billion transistors? Designing for each smaller node has become more expensive, but that’s a very different claim.
This is not a surprise. Why do you think Apple and ARM have been so aggressive about improved performance through increased IPC rather than frequency? Apple A12 is ~4x as fast as A7, but only twice the frequency. The other half of the performance jump is better IPC.
Intel had access to exactly the same info. But they kept up the mad push for higher frequency, apparently learning nothing from the Pentium4...
If you look at the Geekbench scores comparing Whiskey Lake (which has some hardware mitigations) to Ice Lake, an IPC improvement of over 20% is visible over Whiskey Lake. However due to its lower clock rates, the single core performance is 2% lower and multicore 5% higher. Hard to market single digit performance increases when the selling point for Coffee Lake was something along the lines of "100% more cores and 60% more performance!" 5% is hard sell. GPU and platform improvements are easier to make a case for.
> 32 EU parts are listed as "Ice Lake U UHD", while 48 and 64 EU parts are "Ice Lake U Iris Plus" > By increasing the number of execution units from 24 EUs to 64 EUs, Intel is promising over a 2x boost in graphics performance
This is a marketing gimmick then, comparing older non-Iris parts with new Iris parts. If they had compared older Iris parts with new non-Iris parts there would be a reduction in EUs from 48 to 32.
Also key is how much memory is going to be in the the Iris parts. Iris Pro 6200 had 128MB. If these are called "Iris Plus" then the memory could max out at 64MB.
I'd venture it doesn't matter as much. ~60GT/s vs 100GT/s isn't nearly the advantage it used to be (especially considering memory architecture differences between DDR4 and Chrystalwell derivatives), if it even is an advantage at all.
I am wondering about TDP on these parts: At least for Desktop CPU, Intel's TDP is given for the base frequency. Does this apply for these mobile CPU as well ?
This would be interesting seeing how e.g. the i7-1065 G7 has a base frequency of 1.3 Ghz, vs. a boost frequency of 3.9 Ghz.
That it was just refreshed makes the timing a bit strange. Apple is refreshing faster now, but not 3-4 months between a refresh and a major redesign fast. I think they did it because the 15" parts are further off, this would only go in the 13", and they won't have the 13" updated alone.
Which 10nm SKUs suitable for including in Macbook/Air/Pros. The 15" Macbook Pros already supporting 6 and 8 cores i7 so are there equivalent 10nm ICE processors in performance wise can take place inside current 15" Macbook Pros or we have to wait well in 2020 ?
Any vendor could use an i5, disable HT and lower clocks for lower energy, if there was a market for it. Don't think Intel makes any "core"-class dies that don't have the physical assets for HT.
sorry to burst your bubble HStewart . but its tsmc.. not intel.. tsmc may not have the same issues as intel has for the last FOUR years with their node improvments...
"Users might be surprised about this 18% number, given that Intel has historically only provided low single digit boosts in recent generations." No, none of us will be surprised, because none of us will compare the IPC of Ice Lake to the IPC of a CPU series that was released in *2015* (not 2016). We will compare Ice Lake's IPC to that of its direct predecessor, which Intel is reluctant to do.
"The reason for those single digit boosts comes down mainly to small parts of the core microarchitecture being widened." Ice Lake-U almost certainly has single digit extra IPC over Whiskey Lake-U as well. Intel compared their single thread performance (not IPC) only in a single graph, and they reported that Ice Lake-U has a ~4% higher performance than Whiskey Lake-U at 15W. Given their well-known love for skewed scores the actual number must be 2% at best.
The low clocks their 10nm+ node is able to reach has eaten away nearly all the IPC gains of Ice Lake over the direct predecessor CPUs, but given the small IPC gains that have gradually accumulated since Skylake I strongly doubt these gains are more than 8% (which, assuming the actual performance gain is 2%, would imply 6% lower sustained clocks) or 9% tops (for 7% lower sustained clocks).
"It will be interesting to get hold of a system and to actually test/compare with Skylake, especially with all the security patches applied." Er, why on Earth would you want to compare Ice Lake to Skylake?! Did you actually mean the Skylake arch?
You compare Sunny Cove/Icelake to Skylake, because Skylake is the current microarchitecture. Skylake, Kaby Lake, Coffee Lake, Coffee Lake Refresh, and Whiskey Lake (did I forget any?) all have the same microarchitecture, and therefore the same IPC. There are some Meltdown/Spectre Mitigations in Coffee Lake Refresh AFAIK, but otherwise the microarchitecture is untouched.
the time of intels reign is over AMD will be the leader for a few years again now, they have the same momentum intel had with the core 2 release and although intel has some comebacks i believe the advantage AMD has in multicore and design for heterogeneous computing will keep them ahead, have always bought AMD so am pleased that they are back on top again i dont hate intel i just prefer the actual response of amd desktops in the real world, the fact Ryzen is such a good chip just adds the cherry to the cake, awaiting 7nm release to order :)
not a bot.. just can only see intel, and sees intel as the god of the cpu... can do nothing wrong, never makes mistakes, is always right, etc...
he has been asked a few times about the many wrong intel has done over the years.. and he either never replies.. or tries to twist, bend and make up facts to put intel back into the light...
An 18% jump in IPC against... Sky Lake which debut in 2015. This is disappointing and misleading to say the least. Not sure why Anandtech didn't point this out.
Except when you realize IPC has been the same since then. It's clock speeds that have risen. The core architecture is largely the same since Skylake, so an 18% IPC bump over that is an ~18% IPC bump over everything since.
Maybe Asus/ASRock/Gigabyte will make a desktop board supporting Ice Lake mobile processors, like back in 2003, when the Pentium M mobile chips were introduced.
The Yonah/Banias cores were faster/lower power/better than the Prescott Netburst P4 desktop chips that were starting fires at the time.
Intel's partners however have given us some specifications, and it shows in interesting adjustment in the naming scheme. "an" not "in". Intel's partners however have given us some specifications, and it shows an interesting adjustment in the naming scheme.
Intel's partners however have given us some specifications, and it shows in interesting adjustment in the naming scheme. "an" not "in". Intel's partners however have given us some specifications, and it shows an interesting adjustment in the naming scheme.
Intel has done this already - during the Pentium time, when AMD introduced multicore Athlon, Intel came with the completely new Core architecture in low wattage CPUs for mobile, and more than a year later they used Core in desktop and newer node process. Then, it was complicated because of agreements with Rambus, but essentially the process was the same.
I am doubtful about the readiness of Y-series. XPS 13 2-in-1 has been the flagship Y-series device for years but Dell just added fans to its new Gen10 model.
Lukewarm as AMD has no 7nm for mobile for a while but glad to see nice upgrades with IGP. Oh, that i3 is terrible, could have been a quad core with lower boosts or disabled HT, leaving the dual core to a Pentium or Celeron brand.
According the roadmap of intel(include some information by Huawei leaking ),intel will unveils cooper lake-sp in 2019q4 and unveils ice lake-sp in 2020q1 for datacenter. Those production will sell in 2020Q1.
I find it astounding that so many commenters in a site like this do not understand rudimentary concepts. Let’s clarify a few things: IPC = instructions per clock cycle (instructions divided by cycle) Frequency/clock speed = Cycles per second, measured in Hertz (cycles divided by second)
When determining which is faster, you can measure either: 1) Instructions processed in the same amount of time (where a higher number of instructions is faster) In this case, we can calculate the instructions through: X instructions = instructions/cycle * cycles/second * seconds 2) The time it takes to process the same set of instructions (where a lower duration is faster) X seconds = instructions / (instructions/cycle * cycles/second)
Of course, since the above are being measured per core, the other variable you can modify is the number of cores: X instructions = (instructions/cycle * cycles/second * seconds)/core * y cores
Given the geek bench scores posted earlier, we have the following: Single core Whiskey Lake i7-8665u, cTDP up to 2.1GHz, boost 4.8GHz - 5341 Single core ICL i7-1065G7 cTDP up to 1.5GHz, boost 3.9GHz - 5234 Multicore Whiskey Lake - 16485 Multicore ICL - 17330
Single core IPC based on base clocks: ICL IPC is 37% higher Single core IPC based on boost clocks: ICL IPC is 21% higher Raw single core performance: ICL is 2% slower Raw multicore performance: ICL is 5% faster
I don’t have the boost clock based on active cores table for ICL, so we can’t accurately extrapolate the data there. But based on base clocks, there is a 5% performance increase with 30% lower frequencies, meaning there is a remote possibility that clock for clock it’s 50% faster in multicore scenarios.
In reality, the figures obtained are a combination of the chips running at different frequencies and for the short duration that the Geekbench tests run, the chips will primarily run close to the top rated boost clock speeds.
The IPC number is higher than what Intel is claiming because the work load that the CPU architects chose to model while designing the microarchitecture is different from what Geekbench is using. Based on the figures from the computed IPC based on boost clocks, they are similar though. The modifications they make to the microarchitecture are informed by the work loads they choose and there may be regressions in some work loads if they consider the improvements in other work loads to be more valuable.
Note: This only applies when comparing the same set of instructions. If the instruction set architecture (or even just the instructions used, like vectorized vs scalar instructions) is different, then the number of instructions processed is not equal to the amount of work done. E.g. if in one chip with one ISA, the same logical operation requires 1 instruction to perform that takes 6 cycles (IPC is 1/6), while another chip with a different ISA requires 3 instructions that take 2 cycles each (IPC is 1/2), the second chip has a higher IPC but in terms of actual work done, they are identical.
Both Intel and AMD implement the same ISA (with a few differing instructions) and thus can be directly compared for the most part. Recall that when AMD and Intel were competitive in the early 2000’s, Intel heavily marketed their higher frequencies and AMD named their Athlon XPs with names that didn’t represent the CPU frequencies because they had lower frequencies but higher IPC. AMD provided better value then, and with Ryzen 3, will appear to do the same again for the desktop market.
To generalize the Intel chips we’ve seen over the past few years we have: Sandy Bridge Ivy Bridge = Sandy Bridge + better GPU + 22nm Haswell Broadwell = Haswell + 14nm Skylake Kaby Lake = Skylake with higher frequencies Coffee Lake = Skylake with higher frequencies, with more cores Coffee Lake Refresh = Skylake with higher frequencies, with more cores, with more cores
What Intel was targeting to have is: Sandy Bridge Ivy Bridge = Sandy Bridge + better GPU + 22nm Haswell Broadwell = Haswell + 14nm Skylake Cannon Lake = Skylake + AVX512 + new memory module + 10nm Ice Lake
In the amount of time between Skylake and Ice Lake, Intel could have finished development and begun shipment of yet another microarchitecture (beyond Ice Lake). However, because Intel’s microarchitectures are tightly coupled with a process node and Ice Lake was designed with the 10nm libraries, Intel could not produce chips with a new microarchitecture in 14nm. Since they are using the same microarchitecture and, thus, had constant IPC, the only remaining variables they could change to improve performance are: A) the frequency, and B) the number of cores. Kaby Lake modified variable A, Coffee Lake and Coffee Lake Refresh, modified A and B.
From the power usage perspective, we have the following: 1) shorter amount of time to perform the same work load, given higher frequencies and more cores 2) longer amount of time CPU spends idle thus longer amount of time in low power states 3) higher peak power usage because of more cores running at the same time 4) lower average power usage because CPU is idle longer This plays a large part in Intel’s logic of keeping the TDP number identical. Obviously, if they defined TDP to represent peak power usage, it isn’t possible to keep the same figure if you have more cores running with mostly the same manufacturing node. If a core requires 2W to run, logically, running 2 cores will not also only require 2W at the same node.
Because the 10nm launch is so delayed and we’ve had Coffee Lake, et al to compensate, we have the following effect: die sizes are larger, but wafer sizes are the same. This means fewer chips per wafer, and thus lower capacities. Add to that binning for products like the 9900KS, and we have to go through multiple wafers to win the silicon lottery and get dies that can be placed in halo products. This in turn affects supplies to customers, where Intel wasn’t able to cope with the demand. This has caused Intel to invest in more 14nm fabs. This will likely cause Intel to produce more 14nm products for an even longer duration to make up for the costs of investing into more 14nm fabs.
Once Intel decided it would move away from the Tick-Tock strategy and embrace Process + Architecture + Optimization, they told us that 10nm would have lower performance than 14++ and 10+ would be roughly equivalent to 14++. This appears to be because 14+ and 14++ products allowed high frequencies that 10+ products are compensating for through higher IPCs. It is easy to see why Cannon Lake became a non-starter. The same IPC with lower clocks meant lower performance per core and low yields meant you had to keep the die size tiny (and consequently core count low) to have a viable die. Fewer cores, lower frequencies, and high costs means Cannon Lake was always going to be a disaster by the time we got Coffee Lake. Ice Lake with its higher IPC is able to compensate for the lower frequencies to provide comparable performance. The marketing for Ice Lake is touting Gen11 graphics and improvements to the platform because the CPU performance is essentially identical due to the less mature node.
TLDR IPC gains are not hogwash. Intel is comparing ICL to SKL because Intel has ONLY been shipping SKL variants for 4 years now. In fact, IPC gains are the only reason Intel can afford to ship 10nm products now from a user perspective. ICL will perform slightly slower for single threaded applications because of lower clocks but perform slightly faster for multithreaded workloads because IPC gains compensate for the lower clocks. This doesn’t mean ICL itself underperforms. It’s currently hamstrung by the maturity of the node.
Is it reasonable to count AES and AVX-512 to Ice-lakes advantage? Ice-lake has better AES acceleration (21%ST/ 33%MT improvement in Geekbench compared to i7-8665U). Ice-lake gets a boost in GEMM and FFT from AVX-512 (curiously it performed worse in ST GEMM than i7-8665U). My point is that support for specialised code improves the score of the Ice lake, making a general comparision harder. It is a nice boost in workloads that support it, but those are few and far between at the moment. It is also worth pointng that AVX instructions doesn't run at the same clock speed as integer instructions.
I wouldn't say they're unreasonable if they accelerate common workloads. Intel's been marketing how its specialized instructions improve performance since MMX. AMD's done the same thing with 3DNow! On that note, I don't think Intel's leaning heavily on AVX-512 for their IPC improvement claims. The product brief they released has a disclaimer about AVX lowering clock speeds and people have reported that AVX-512 is causing their CPUs to throttle, slowing down execution of even non-AVX code. If thermal issues are still a problem with AVX-512, they're probably amplified in Ice Lake because of the smaller die sizes. It's harder to dissipate heat when the surface area is smaller. Hopefully Ian can score some samples soon and we'll get to see real results. If anything, cache improvements and a better back end probably account for most of the IPC gains. We've seen what wonders that does for every generation of Apple A chips. Again, I'm eager to see what Ian has to share.
Yes. I guess we just have younger readers and don't understand what is going on with Intel. Aside from that, we haven't seen proper reviews yet and AMD will delay 7nm mobile. IceLake is important, though I don't like specs for the i3 chip with only two cores.
No delay from AMD on laptop chips, but AMD always starts with high end desktop to get the chips into systems quickly. If AMD is planning on Navi going into the 7nm laptop chips for the next generation, making sure that the GPU drivers work will be good for OEMs as well. The question is when AMD will release the new laptop chips. AMD released the 3000 series laptop chips(based on 12nm Zen+ and Vega) in January of 2019. If AMD gets the next generation out in November of 2019, that would be a positive thing.
A big factor that makes the stated numbers more of a challenge is security mitigations, and not having all mitigations in place when Intel stated the +18% IPC figure. This implies that the +18% figure isn't going to be accurate once the mitigations are in place. Most people are not saying that there are no improvements, but are pointing out that the actual IPC improvement may be completely offset by security mitigation numbers combined with lower chip frequencies. To make it even more difficult, Intel is known for setting TDP figures that only apply to the base frequency. If base frequency is only 1.1GHz for a 15W TDP chip, and that 15W is only accurate at the base frequency, then is Intel actually doing well compared to an AMD Ryzen 5 3500U or Ryzen 7 3700U in laptop performance?
You're right that _some_ of the comments were theorizing that the security mitigations account for most of 18% figure. I've previously responded to some of those comments to point out that the Geekbench numbers quoted are from Whiskey Lake and Ice Lake. _Both_ have some security mitigations in hardware. As I mentioned in my post, you'll see an IPC improvement higher than what Intel claimed because of a difference in the workload being tested. I did, however, also mention that there's barely any improvement visible when looking at the raw performance because of the low frequencies. This is why I said the IPC gains are the only reason Intel can afford to ship 10nm products. They can't sell something that performs worse but they also can't market single digit performance improvements and slight regressions so they focus on the GPU and platform improvements.
The main reason I posted anything at all is because it's frustrating to see that people don't have a clue what IPC means. While _some_ of the posts brought up spectre/meltdown/mds, a lot of the others appear to think that each set of chips Intel has released over the past 4 years had IPC gains. I explained that Intel had to keep the IPC constant and rely on the two other factors they could change to improve performance: a) frequency, and b) core count.
In terms of laptop performance, it'll be hard to get a fair comparison regardless because OEMs tend to put AMD chips in cheap laptops with low end specs. Ice Lake does have a process advantage over the mobile Ryzen 3 chips since those are still manufactured at the 12nm node. We'll need to wait for reviews of actual products to see the difference. Moreover, the overall platform is a bigger deal in mobile products because of battery life concerns. In terms of laptop energy consumption, it isn't just the CPU that matters. Intel has integrated the functionality of more chips into its new PCH. This means that there will be fewer chips, lower overall power draw, smaller motherboard footprints, and lower costs for manufacturers. Integrating WiFi and Thunderbolt into the PCH alone would save a considerable amount of power.
Yup. I'm surprised so many in the AT audience don't understand this. IPC has been the same since Skylake, it's the clock speeds/boosts that have risen, so 18% over Skylake is very nearly 18% over everything since on Instructions Per Clock.
Here we will control on the most proficient method to introduce stock ROM on BLU Studio Mega 2018. The guide is straightforward and simple to pursue. You can watch our total video direct. https://blucellphones.us/download-blu-usb-drivers-...
decide when and where to apply charges. Weebly gives and keeps up current city, state and region level assessment rates for the United States and Canada. For every single other nation, charge rates can be characterized by the vender.
decide when and where to apply charges. Weebly gives and keeps up current city, state and region level assessment rates for the United States and Canada. For every single other nation, charge rates can be characterized by the vender.
decide when and where to apply charges. Weebly gives and keeps up current city, state and region level assessment rates for the United States and Canada. For every single other nation, charge rates can be characterized by the vender.
"My grandma made jugs of this," says Tara, "and my sister and I, joined by our canine, would bounce from house to house in Belgaum gifting them to her companions. Consequently, they gave us their custom made summer specialities like Coorgi severe orange (kaipuli) squash, sharp mango pickle, or Goan dried fish pickle." https://pandareviewz.com/sindhi-food-cuisine/
the administrators. By making sense of how to manage your time effectively, handle stress in a beneficial way, and keep up a calm, elevating angle, you'll have the alternative to rest better around night time. www.sleepnpeace.com
Ah, they aren't actually shipping these 10nm laptop chips just yet. However, this does mean that all those "leaks" of chips with designations like i7-10700F are fakes, since the naming scheme for the 10th generation is keeping things at four-digit numbers!
A portion of the low, low costs you see on the greeting page for these web facilitating administrations are early on offers. At the point when the special time frame closes, the customary costs can be essentially higher. One essential special case is Complete Internet Solutions, which guarantees, "When you sign up your cost will never go up." <a href="https://wphostingbasics.com/bluehost-high-traffic-... traffic site hosting</a>
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
190 Comments
Back to Article
Teckk - Wednesday, May 29, 2019 - link
Will be good to have a significant jump in performance + jump in battery life.Also, wasn't it data center first with new nodes - IceLake Xeons followed by the mobile and desktop or that's not for this gen products?
goatfajitas - Wednesday, May 29, 2019 - link
No, I dont think Intel has done that at all, at least not in the past 15 years or so. It's generally laptop first, then desktop, then Xeons.jospoortvliet - Wednesday, May 29, 2019 - link
Yup. And imagine they hadn’t had this big 10nm Delay - a 18% IPC improvement would have made ZEN2 a whole lot less interesting. Alas, it will be another year before these reach desktops and by that time, amd has ZEN2+ out ;-)goatfajitas - Wednesday, May 29, 2019 - link
"Milk it" for a dozen years and this is what you get LOLHStewart - Wednesday, May 29, 2019 - link
One thing that Intel is doing smart is learning from the 10nm delay, they separate the process from the architexture. In the market today desktop is minor market, just a bunch of gaming kids. Mobile is the primary market nowKorguz - Wednesday, May 29, 2019 - link
time will yell if intel learned anything....Korguz - Wednesday, May 29, 2019 - link
grrr tell.... not yell...Father Time - Wednesday, May 29, 2019 - link
INTEL IS LEARNING!!!! THEY'RE LEARNING AND GROWING STRONGER!!!Korguz - Wednesday, May 29, 2019 - link
yea right.. not likelysurt - Friday, May 31, 2019 - link
That is the funniest thing I've read on the internet this month, maybe this year. Well done.adlotery - Monday, July 1, 2019 - link
We are consistently refreshing all these, morning, night and night results on our site. You don't have to meander around for getting the Sambad lottery result today, you are most welcome here to check your lottery result on our site effectively. https://todaysambadlottery.in/desiredderised - Saturday, July 20, 2019 - link
ha! thx!desiredderised - Saturday, July 20, 2019 - link
Yeh! Just like your mom!ajp_anton - Wednesday, May 29, 2019 - link
That's exactly what they haven't done. They've continued to push 14nm revisions of Skylake over and over instead of backporting Ice Lake / Sunny Cove to 14nm. That's why Ice Lake was delayed so much, it had to wait for 10nm.name99 - Wednesday, May 29, 2019 - link
So they SAY.Intel says a lot of things, most of them contradictory (right now they are telling us BOTH that there will be speed optimizations on 10nm, AND that there will be a rapid transition to 7nm ?!?)
I’ll believe this story when I see it. Given how slowly Intel changes, even if they did actually agree to this, it may take 5 years before it’s implemented.
goatfajitas - Wednesday, May 29, 2019 - link
Yup, as with anything, it is nothing until its released and retail units (not engineering samples) are independently tested. Then an only then is it real.0ldman79 - Thursday, May 30, 2019 - link
This is a very good point.Mobile is where it is at right now.
It looks like Sunny Cove arch is going to nearly match desktop performance, at least in short bursts, in spite of the lower clocks.
I don't think that's going to hold true on extended runs. Encode a short video and Ice Lake will probably kick ass. Encode a longer video and this likely won't be the case, it'll hit that 25w TDP and drop off.
wanderer66 - Thursday, June 13, 2019 - link
Their bread and butter is the giant-scale cloud and enterprise market. ironically, they are among the most sensitive to overall power consumption due to scale. Intel began the effort to switch to developing more power efficient cores more than 10 years ago however. They're not the leader because ARM already was, and started from the other direction. You may also well notice that neither has done particularly well moving into each others architectural strongholds.evernessince - Wednesday, May 29, 2019 - link
People need to pay attention to that IPC number. That's only in comparison to older Intel processors, not current gen. They got a few people here clearly with that trick.Siats - Thursday, May 30, 2019 - link
Skylake is current gen in regards to IPC, 7th to 9th gen are just iterations on it with no IPC gains at all.danielfranklin - Thursday, May 30, 2019 - link
Exactly, IPC at the moment is based off skylake CPU cores.All we have got is more cores bursting to high real TDPs.
Remember that Intel basically skipped/ditched their traditional “Tick” ; first gen 10nm parts.
Targon - Friday, May 31, 2019 - link
This +18% IPC doesn't include the security mitigations, and note the clock speed. From Intel, 15W TDP is for the base speed, which is quite a bit lower than the 14nm chips.Rickjason - Saturday, June 1, 2019 - link
This 10nm delay is pure incompetence. Would this have happened under Moore or Grove? Intel, instead of innovating, keeps on making minor improvements on old designs. They really need a new management team with some imagination.Korguz - Saturday, June 1, 2019 - link
Rickjason part of the lack of innovation.. could be due to the lack of competition, just like back before the athlon 64 came out, and caught intel with its pants down.. just like with Zen :-)Korguz - Saturday, June 1, 2019 - link
oh.,. and the fact.. that intel just cares about profits and lining its pockets :-)HStewart - Wednesday, May 29, 2019 - link
Usually Intel releases new technology in Xeon's first for example AVX 512 and since most of industry is going mobile now - then it logic to do mobile first. I would expect the Ice Lake based Xeons to have new technology components in it.Korguz - Wednesday, May 29, 2019 - link
most of the industry is goimh mobile ?? yea sure.. prove it... no one i know is buy laptops... they are still looking at desktops.. and for the moment.. they are looking at zen 2.. not intel...1_rick - Wednesday, May 29, 2019 - link
Corporations seem to be heavily transitioning to laptops for most users, however.Korguz - Wednesday, May 29, 2019 - link
corps.. yes.. makes sense.. so they can then have then bring their work home with them :-) for general public.. im not so sure...PeachNCream - Wednesday, May 29, 2019 - link
Its true of the general public too. Walk into a retail store that has a history of selling computers and you'll find lots of laptops on display with maybe one desktop PC available but not even out on a shelf. Desktops are still around mostly as gaming systems because of upgrading and bling. Apart from that there are still heavy lifting workstations and then mini PCs (made largely of laptop-class components). Desktops are few and far between which is why there is now a large price premium for components in order to keep margins a shrinking segment high enough to continue pulling profits as sales volumes taper off.Korguz - Wednesday, May 29, 2019 - link
thats not the case here.. the stores may be like that.. but no one i know is interested in laptops... i havent heard any of their friends say they want a laptop.. i even keep getting asked about desktops, and what they should get .... maybe its different market needs for the areas ?drothgery - Thursday, May 30, 2019 - link
Reality is almost twice as many laptops & 2-in-1s are sold these days as desktops, and the ratio is going to keep going farther and farther that way. You and your friends probably have similar work or interests (PC gaming is the most likely, since you're on a hardware enthusiast site) that makes desktops more useful, but the people you know are not a representative sample of the broader market.Korguz - Friday, May 31, 2019 - link
drothgery nope.. various usage.. games, general web browsing, music listening... they just dont have any interest in laptops for some reason... i am yes. but them.. they dont pay attention to hardware news... they are not.. but still.. a few here seem to think every one wants laptops now... which im trying to say... isnt the case ...Thud2 - Saturday, June 1, 2019 - link
You keep using elipses. I do not think it means what you think it means.Striderevil - Thursday, May 30, 2019 - link
No not really. Though laptops provide the convenience, look at any office. Most are connected through desktops. Desktops provide the convenience of easy maintenance, lower costs and upgradability while a laptop is mostly obsolete by 5 years and cannot be upgraded. Most developers, graphic designers and marketing mostly use a desktop. The only profession where a laptop is popular is among students and journalists.simonpschmitt - Thursday, May 30, 2019 - link
This is not true for all companys I worked for.I work in pharmaceutical R&D and in chemical R&D before thet and litterally every (non production) employee gets a laptop. From assistants over researchers to the CEO.
And then some of the scientists have a desktop/workstation for CAD or other demanding work. In addition to their laptop.
Gigaplex - Friday, May 31, 2019 - link
Every office I've looked at replaces their machines within 5 years anyway, they don't bother with upgrading desktops. I'm a software developer and have usually worked on a desktop, but on my last hardware refresh I was moved to a laptop.RSAUser - Saturday, June 1, 2019 - link
In software here, organization of about 100 devs this branch, pretty much 90%+ are on laptops with the others using both.Anything compute intensive is thrown at a server farm to compile/execute, an IDE doesn't need much performance.
pk1489 - Thursday, May 30, 2019 - link
According the roadmap of intel(include some information by Huawei leaking ),intel will unveils cooper lake-sp in 2019q4 and unveils ice lake-sp in 2020q1 for datacenter. Those production will sell in 2020Q1.0ldman79 - Thursday, May 30, 2019 - link
I think it's mostly been desktop, laptop, server. Broadwell flipped the script, pretty certain Skylake was desktop first, Haswell too. Sandy Bridge was Jan 2011, mobiles didn't come out until Sept 2011. Ivy Bridge was released in April 2012, both laptop and desktop.Santoval - Wednesday, May 29, 2019 - link
Based on their own graphs only the 64-core iGPU will provide a significant jump in GPU performance. Due to the low clocks the CPU cores will struggle to provide more than 2% extra performance (over Whiskey Lake-U) at best. To be clear the +18% IPC is over Skylake CPUs, which were released in 2015, not over the current Whiskey/Amber Lake U/Y CPUs.Oh, there will also be quite a jump in encoding performance, for those who prefer to encode video in hardware (for me that's very important). Decoding videos should also become even smoother, lowering CPU utilization further (no word about AV1 decoding yet though). I don't care about the faster wireless because I don't throw LAN parties, while the benefit of the AI speedup is unclear.
s.yu - Wednesday, May 29, 2019 - link
This is as I read it too, there's again minimal improvement generation over generation meaning this will do nothing to stop AMD from surpassing them.HStewart - Wednesday, May 29, 2019 - link
It funny how people base there facts on desktop cpu's but don't realize these are Laptop chips.Korguz - Wednesday, May 29, 2019 - link
its funnier how you are still pushing intel when its pretty clear that intel.. may NOT be the performance leader anymore... and baseing things on mobile.. is still valid as it could translate to desktop performace still... but we will have to wait to see to be sureRanger90125 - Thursday, May 30, 2019 - link
Your perpetual pro Intel posts never fail to raise a smile. It very funny....Manch - Thursday, May 30, 2019 - link
Half the market is still desktops and they still maintain a slight lead over laptops. Also desktops are on the rise bc performance is finally on the rise.ajp_anton - Wednesday, May 29, 2019 - link
Whiskey/Amber/Coffee/Kaby/whatever Lakes are all Skylake, so identical IPC (excluding security fixes in hardware, but that should really be treated separately from the Skylake architecture's IPC).Where did you find graphs that compare anything other than the 64-core version? Which BTW "based on their own graphs" seems to be about 60-70% faster than Whiskey Lake-U.
name99 - Wednesday, May 29, 2019 - link
(a) When full Spectre mitigation costs around 15%, that’s not a trivial point in terms of cross-device comparisons..(b) The IPC comparison is interesting in terms of suggesting Intel’s future, but if you are buying a chip, what matters is performance, not how it’s split between IPC and frequency.
AVX512 may matter for future apps, but it’s not unreasonable to point out that these are NOT really faster chips (for most single threaded performance) no matter what Intel says.
jospoortvliet - Thursday, May 30, 2019 - link
The Spectre impact depends on the benchmark, certainly there is a 15% worst case but over a big range of benches the diff is on average more like 3-5%. Certainly on spec which is probably the benchmark intel got the 18% from.Targon - Friday, May 31, 2019 - link
Meltdown, Specter, Spoiler, Zombieload, the list goes on. Mitigations for all of these add up, probably more than just 18% worth.1_rick - Wednesday, May 29, 2019 - link
"only the 64-core iGPU will provide a significant jump in GPU performance"" To be clear the +18% IPC is over Skylake CPUs, which were released in 2015, not over the current Whiskey/Amber Lake U/Y CPUs."
Yeah, there doesn't seem to be enough understanding of this. Intel's claiming perf improvements over 4-generation-old processors is a bit disingenuous.
Also, for those relishing the idea that the Intel iGPUs may now be better than the Ryzen ones, remember that AMD can probably do the same thing and add more cores to THEIR APUs as well. (Whether they will or not is another matter, of course.)
jospoortvliet - Thursday, May 30, 2019 - link
Note that Skylake was their last real architecture the rest was just process improvements. So the IPC of whiskey lake and Skylake are the same...deil - Thursday, May 30, 2019 - link
small first -> better yeld.bigger when process matures so they can keep costs down, and squash bugs/issues before high expectation xeons appear...
BigMamaInHouse - Wednesday, May 29, 2019 - link
Intel already has i7-8569U [4.70 GHz boost] with Iris Plus Graphics 655[921.6 GFLOPS ] with same 28W TDP, so how much better the 10nm part looks like now[at least from 14nm++ down to 10nm]?abufrejoval - Wednesday, May 29, 2019 - link
In terms of absolute top compute power, probably no difference, neither for CPU nor GPU.It's all about the Watts and the ability to sustain that max power closer to 15 Watts then 28 Watts, which makes all the difference in the world running on battery and for noise/form-factor.
And btw. good luck finding the current Iris Plus parts in anything but
* insanely priced Macs
* astonishingly priced Intel NUCs
Astonishing, because Intel doesn't *charge* extra for the much bigger GPU and the eDRAM, they just won't *sell* it to anyone but Apple and themselves.
Somehow Intel must have decided
1. We don't charge *ever* for iGPUs (actual price difference seems around $5)
2. We don't make enough profit selling 48 or 72 EU parts to the open market under rule 1, so we won't
BigMamaInHouse - Wednesday, May 29, 2019 - link
Edit:Dell XPS 13 7390 Ice Lake Core i7-1065 G7 vs i7-8665U in Geekbench
https://browser.geekbench.com/v4/cpu/compare/13233...
IMO AMD's Zen2.0 with 7nm shrink looks better improvement.
HStewart - Wednesday, May 29, 2019 - link
I think you are comparing 8th gen U vs Ice Lake Y - it appears on spec that the new XPS 13 2in1 is Y based processor and not the Ice Lake U processor. I have original XPS 13 2in1 and low based ghz and max ghz sounds like this is Y processor and not U processor.BigMamaInHouse - Wednesday, May 29, 2019 - link
it's 4C/8T 15w TDP same as i7-8665Uhttps://www.notebookcheck.net/Dell-XPS-13-7390-2-i...
HStewart - Wednesday, May 29, 2019 - link
My thought is that it could be Y as existing XPS 13 2in1, just Intel has increase the core count. the specs indicated 9W, 15W and 28W but only 4 cpu's listed. This likely means more processors are to come and not a guarantee that that one in XPS 13 2in1 is U processor. Also 28W version should be a lot faster because of higher watts.HStewart - Wednesday, May 29, 2019 - link
It hard to tell now what the exact performance will be - but Windows Central states that new Dell XPS 13 2in1 could be best convertible ultrabook of 2019. Notebook check did indicates that it is launch on July 10 - you know about 40 days from now.One thing that is confused only 4 models list here, 11 models coming and 3 different watts This is summer 2019 release and what is being release for holidays. Intel is probably waiting to see how AMD responds and come back hitting.
https://www.windowscentral.com/xps-13-7390-compute...
jospoortvliet - Thursday, May 30, 2019 - link
Well intel can wait a long time, AMD lacks a compelling solution that competes with this in the (ultra) portable market... it will take a while until zen2 and Navi come to mobile, long enough perhaps that they even end up competing with whatever intel does next. AMD clearly focuses on enthusiast desktop & server for now, where they can win easily. Sadly for them laptops are a huge market... though of course there is more $$$ in servers.Targon - Friday, May 31, 2019 - link
Take a look at the clock speeds on these chips. Yes, turbo goes higher, but at 15W, that is base clocks, and turbo we don't know how much power or heat gets generated. AMD SHOULD move up the next generation laptop chips by a few months if possible, because a 7nm Zen2 based laptop chip would destroy these new Intel 10nm laptop chips.dromoxen - Monday, June 10, 2019 - link
there is no such thing as an AMD NUC .. (mini-pc) believ me I've looked . A whole segment of the market going begging Under 15w from AMD Mini-Pc would be great ... maybe some day .. when their chip supply improves?RSAUser - Saturday, June 1, 2019 - link
Wasn't the issue with AMD chips something along the lines of not properly ramping up or down, wasting a lot of battery?Once fixed playing field is even at worst, while AMD is releasing an improved APU based on Navi, just no die shrink.
Santoval - Wednesday, May 29, 2019 - link
The CPU clocks are going to be quite lower than the 14nm++ equivalents, due to persisting thermal issues and low yields. I have no idea what that means exactly for the iGPU clocks, but I would expect quite lower iGPU clocks as well, or at least much shorter sustained iGPU boost clocks.Gen11's design is somewhat different though, so we will need to wait for devices to be released to find out how the iGPUs will perform (we already know, based on Intel's own graphs, that Whiskey Lake-U is going to have crappy performance - at best 2% faster than Whiskey Lake-U . Their thermal issues also had to force them to blow up the TDP of the Y parts from 4.5W to 9 - 12 W, which is staggeringly high and probably precludes fanless designs, i.e the designs the Y CPUs are actually meant for...).
Santoval - Wednesday, May 29, 2019 - link
edit : I meant the Sunny Cove cores are going to have crappy performance, not Ice Lake-U by and large.firewolfsm - Wednesday, May 29, 2019 - link
The TDP growth from 5->9W for the Y parts can mostly be explained by integrating wifi and thunderbolt 3, which are worth 2W each at least. The cores themselves are most likely being allocated slightly less power than before, and system power consumption will improve.s.yu - Wednesday, May 29, 2019 - link
Didn't realize Thunderbolt needed so much power.HStewart - Wednesday, May 29, 2019 - link
It also could be related to fast iGPU and also does anything stated there is quad core Y Part. The XPS 13 2in1 previous version was Y part.FreckledTrout - Wednesday, May 29, 2019 - link
A decent improvement on the graphics front however just barely beating AMD's 3700U isn't that impressive. I mean we are talking a die shrink so there is a lot more die space to play with and graphics should scale out nicely with more graphics cores. AMD's first 7nm mobile chip to replace the 3700U should surpass this 10nm chip rather quickly.lightningz71 - Wednesday, May 29, 2019 - link
I wouldn't say that AMD will easily surpass this. The jump to 7nm for AMD's APUs will be an interesting one. It appears that AMD is currently content to have their mobile chips be based on mature nodes and aimed at a value position in the market. With that in mind, I suspect that Renoir, which is the code name for the next gen APU from AMD, will be based on the current 7nm process, will likely continue to be a monolithic die, will likely retain the 4 core, single CCX arrangement that they currently use, will likely have 1/2 the L3 of their existing 7nm CCX design (so, 8MB), and will likely have a NAVI based iGPU section, probably with 20 CUs (more CUs running at lower speeds can be more power efficient). The big gain for AMD will be that NAVI is expected to be more efficient with memory bandwidth (new caching structure, supports advanced features that save on bandwidth usage) and have support for faster DDR4 (though, I don't imagine that we'll be seeing anything much faster than DDR4-3200 in an SO-DIMM in wide usage by next year). I don't expect that AMD will have a memory controller that can support 4 X 32 bit LPDDR4X-3733, so will be at a bandwidth disadvantage.The one possibility that AMD has to surpass Intel here would be for them to leverage their experience with HBM and have Renoir be a design that is capable of being either independently packaged for lower cost applications, or being mounted on an Emib/MCM package and support a "back channel" connection to a 4GB HBM stack. A 20CU Navi with 4GB of dedicated HBM with appropriate clocks would likely be firmly in GT 1650/1660 territory, which would be just fine for 1080p gaming. That seems like a reasonable sweet spot for mobile devices and something that could also see reuse as a desktop APU for Home Theater machines that live in mini-ITX cases.
Zizy - Wednesday, May 29, 2019 - link
As Intel's upcoming solution is barely beating existing 3700U devices (in Intel tests), AMD will surely regain lead with 4700U even if it remains on the 12nm process. If it doesn't actually lead already in independent tests.Switching to 7nm is fairly likely next year (keeping just 2C/3CU cheap parts on 12nm) and will greatly surpass IceLake. But by then Intel will finally manage their dGPU. I wonder how this one ends up.
So far, GPU engineering hasn't shown anything wow, but with their new push in marketing they might launch something interesting. Or at least their PowerPoint will say so.
Hifihedgehog - Wednesday, May 29, 2019 - link
Navi + 7nm will likely smoke these Ice Lake parts. Someone has done some calculations comparing Ice Lake to Zen 2 in Geekbench and the IPC advantage is more like 6%, not the 18% that Intel was touting. They must have included AVX-512 (not just integer like AMD did) in the mix to give the appearance of extremely good IPC.Hifihedgehog - Wednesday, May 29, 2019 - link
EDIT: One of the Spec tests that Intel ran uses AVX-512. Not much out there utilizes it so that is very misleading on their part.Santoval - Wednesday, May 29, 2019 - link
Intel reported in a graph that Ice Lake-U has a 4% higher single thread performance than Whiskey Lake-U, with both at 15W. They did not compare them directly, because apparently it wouldn't look good. They set Broadwell at 1.00 performance, reporting ~1.43 for Whiskey Lake-U and ~1.47 for Ice Lake-U.What do you think the real number is? I was thinking ~2%, but if they employed AVX-512 to skew the score the actual gains might be in the 0% to 1% range. Basically the question is if their 10nm+ node's very low clocks ate *all* their IPC gains or if it left a few scraps of it. Which is why their main presentation had big graphs shouting about x2 iGPU performance(!), x2.5 AI performance(!) and ... 3 times higher wi-fi "performance"(!), with no word about CPU performance.
I smell a fiasco slowly brewing...
Santoval - Wednesday, May 29, 2019 - link
That +18% IPC was over the (original) Skylake that was released way back in 2015, not over Intel's current CPUs (i.e. since we talking about a TDP of 15W over Whiskey Lake-U). These are technically still based on the Skylake arch but have been optimized and refined multiple times, are fabbed on newer variants of 14nm etc etc Intel did not report the IPC gain over Whiskey Lake because it is obviously very low, and surely single digit. It is difficult to asses it because Intel did not disclose base clocks, but it should be in the 6 to 8% range.ajp_anton - Wednesday, May 29, 2019 - link
Whiskey Lake has identical IPC to Skylake because it *is* Skylake. It has some security fixes, but other than that, they are the same. Faster RAM might give the illusion of slightly higher "IPC", but the only real improvement is the improved manufacturing leading to higher clocks. Actually the clocks have been improved so much that since Skylake we have pretty much had the biggest generation-over-generation performance improvements since Sandy. At least running stock clocks.Korguz - Wednesday, May 29, 2019 - link
ajp_anton.. tell that to HStewart.. he has been saying for a while now that what intel is realeasing now.. is a new architeCture.. when all it is.. is yet another update/refresh...HStewart - Wednesday, May 29, 2019 - link
Look at architexture on chip Korguz, and tell me that it not different than Skylake, it has more ALUs and other logic, larger cache and besides that also AVX-512 and other enhancements. It would be pure stupidly to say it is update/refresh on skylake. Of course you maybe talking about WhiskeyLake and not Ice Lake which I referring to.One difference that I just realize was started on my Dell XPS 15 2in1 Kaby Lake G and available in these new processors in Dynamic Tuning, plus there are other aspects but core IPC performance of Skylake series is same - but they adjusted parts of it.
But I think you were confused between IceLake and other processors - yes we all don't live in desktop line and who knows why Intel calls new desktops 10 series on 14nm.
Korguz - Wednesday, May 29, 2019 - link
HStewart.. 1st.. do you NOT know how to spell ??? its architeCture... NOT architexture.. for some one who says he has been using computers for as long as you " say " you have.. you would think you would know how to spell it CORRECTLY.2nd.. it ISNT'T different then skylake.. its just a refresh/update to an EXISTING architeCture that intel has been feeding us for 6 or 7 YEARS now
" it has more ALUs and other logic, larger cache and besides that also AVX-512 and other enhancements. " what do you think and update/refresh is??? Amd going from bulldozer to zen
THAT is a new architeCture... what intel is doing is NOT but you are too much of an intel fanboy to realize this. you keep believing intels bs and lies.. wake up HStewart.. stop being the fanboy you are.. and open your eyes
catavalon21 - Wednesday, May 29, 2019 - link
Amused at your response to HStewart for misspelling a word, then you proceed to follow with two misspelled words of your own (or one with another wrong word). Spelled correctly, the second paragraph would start (capitalizing words that are corrected) ...it ISN'T different THAN skylake...Glass houses and all
Korguz - Wednesday, May 29, 2019 - link
catavalon21 the joy of typing on my phone.. vs my keyboard.. but if you look at ALL of HStewarts posts on here.. he ALWAYS spells architeCture wrong... ALWAYS, thats the difference...im sure you have spelled words wrong ( like any one else ) when typing on a phone.. vs keyboard :-)
catavalon21 - Thursday, May 30, 2019 - link
Sure, I also make tons of typing errors on a full sized keyboard. I type poorly and often do a crummy job of proofreading. I also know that for many contributors on AT, English is not their first language, and try to keep that in mind when the tendency to get picky hits me.phoenix_rizzen - Thursday, May 30, 2019 - link
All the Lakes are the same basic architecture with refinements and enhancements and the of new feature. It's pretty much advertised in the name.The *wells were based on the same architecture, which changed going into the *Lakes.
No different than how AMD iterated over the Bulldozer architecture all the way through Excavator/Kaveri, then switched to Zen. Then they iterated over Zen with Zen+. Now they have a new architecture with Zen 2.
Maybe next year Intel will have a new architecture to base their chips on that's as different to Skylake as Skylake was to Broadwell. That's when the Faveros and EMIB stuff should get interesting.
phoenix_rizzen - Thursday, May 30, 2019 - link
If you think back to "the old days", an existing architecture was used across two nodes, then a new architecture was released on the new node (tick-tock setup).Skylake was the new architecture on an existing node. All the Lakes have been iterations over slightly tweaked optimisations of that node. And nope they finally get the existing architecture running on a new node. Tick-tick-tick-tick-tick-tock.
Next year will be the new architecture (tick) on 10 nm. There may be more ticks on 10 nm while they prepare 7 nm. Then the tock will happen on 7 nm. And the prices will repeat.
phoenix_rizzen - Thursday, May 30, 2019 - link
Urgh, typing on phones without an edit function sucks. Nope -> now. Prices -> process. Probably other typos too.ajp_anton - Thursday, May 30, 2019 - link
Yes, Ivy was a refinement of Sandy, etc. But all of the lakes after Skylake are not even refinements. They have fixes for security and maybe other bugs, but they are all Skylake. Intel changed the whole naming scheme because of this - one name for the chip, one name for the CPU core architecture (Ice Lake now has Sunny Cove cores).Kaby Lake, Coffee Lake, Amber Lake, Whiskey Lake all have Skylake cores. No tweaks, no real improvements, only bugfixed Skylake cores with identical performance. The only exception to this is Cannon Lake, which actually did have a new core (though it's the same kind of tweak that Ivy and Broadwell were).
Qasar - Thursday, May 30, 2019 - link
HStewart, IF this really was a new architecture, then WHY is intel THEMSELVES calling this 10th gen ?? if it was a new architecture like you wish it was... then wouldn't it be called " intel gen 1 10nm ( insert new code name here ) ??? this is just an update/refresh that intel has been using as the base for their cpus for the last few years as korguz has said. Zen 2, is just an update/refresh of Zen+, which is an update/refresh of the base architecture that was Zen, plan and simple... lets see you try to spin this to make your intel look better then it is.ajp_anton - Thursday, May 30, 2019 - link
I never mentioned Ice Lake, why did you think I did?Ice Lake is finally new. Not new as in being a departure of the "Core" architecture, but it's finally something new.
Skylake, Kaby Lake, Coffee Lake, Amber Lake and Whiskey Lake all use the same Skylake core with identical IPC. Only bugs have been fixed, which may have impacted performance *later, if they are being fixed in software*, but the base IPC is identical. Cannon Lake was the tweak to Skylake that previous tweaks have been (Ivy Bridge, Broadwell), but it was never really released. Ice Lake (using Sunny Cove cores) is supposed to be a new architecture (still within Core), as Haswell and Skylake were.
schujj07 - Wednesday, May 29, 2019 - link
When you figure that compared to Sky Lake the Zen+ was 3-5% lower in IPC that would put Zen 2 at 10-12% higher IPC than Sky Lake, if the quoted 15% IPC increase is true. Therefore if Ice Lake has an 18% IPC increase it will be about 6% higher IPC compared to Zen 2. None of this will be known until systems are out. Also Zen 2 won't be doing battle with Intel for a while since Ice Lake is only in laptops and Zen 2 is only desktop.Meteor2 - Thursday, June 6, 2019 - link
Been reading this whole thread waiting for this analysis. Thank you :-)just4U - Wednesday, May 29, 2019 - link
Well.. Amd is getting some huge play from the OEMs for gaming laptops over the next year.. so you can bet they have a few innovations which are pulling those OEMs over. Supposedly going to see some fairly high end laptops with Amd inside.Still any improvement on graphics by intel is always welcome news.
Targon - Friday, May 31, 2019 - link
I just hope all of these systems with ONLY a 128GB SSD for storage go away, because your typical consumer has no clue, and then freaks out when their storage fills up within a month of purchase due to pictures, music, or whatever.Targon - Friday, May 31, 2019 - link
Desktop chips can be installed on release day, so that means you will see consumers with the new generation on day one, talking about the performance improvements. Laptop chips will take a minimum of three months to go from sending them to OEM laptop makers to QA, BIOS adjustments, cooling solutions, etc. Then another three months before the shipping products actually get into the hands of consumers. Even if OEMs start shipping laptops on launch day, it will still take three months before consumers will be able to get them.Since desktop chips can also go into existing motherboards, you will see a lot of upgrades as well from first to third generation. The marketing splash of enthusiasts talking about the new chips does help drive sales in laptops, but there will always be a bit of a delay.
You also have the effect of Navi...the next generation laptop chips from AMD will probably have Navi instead of Vega graphics, so working the kinks out of the drivers for an extra 3-5 months will make for a better end-user experience.
With the 7nm process for higher clock speeds, the +13-15% IPC, and Windows support being more mature, I'd say that AMD has a good chance to beat Intel in laptop performance as well. Intel 10nm production is going to remain low, while TSMC 7nm is going to be able to churn out chips for AMD in much higher numbers.
tkSteveFOX - Wednesday, May 29, 2019 - link
Took Intel long enough, but now Qualcomm Snapdragon 8cx is coming with a 5G modem for always-on computing.Intel will be tough to beat that unless their Ice Lake chips are cheaper than Qualcomm's.
Gondalf - Wednesday, May 29, 2019 - link
What is the relevance to speak about a Qualcomm SOC unable to run native x86 applications?Apples and oranges.
Snapdragon 8cx is only an exercise of a secondary Qualcomm team with a bad future ahead
ajp_anton - Wednesday, May 29, 2019 - link
It can run x86. It's emulation, so a bit slower, but it can. And the performance is impressive considering it is emulation. That's how it starts, with lower performance but "everything works". Once (or rather if...) it gets momentum, more applications might get native Arm64 versions.1_rick - Wednesday, May 29, 2019 - link
"It's emulation, so a bit slower, but it can."Apparently it's more like "It's emulation, so a LOT slower, but it can", from the previews so far.
bigvlada - Thursday, May 30, 2019 - link
You might want to check a few articles about the previous attempts, DEC Alpha and PowerPC and their's x86 emulation on paper and in real life.Meteor2 - Thursday, June 6, 2019 - link
Why? I'll concern myself with the performance of modern x86 emulation, not times past.abufrejoval - Wednesday, May 29, 2019 - link
Not quite the same market, but yes I agree and I'd love to get one of those 8cx Lenovos into my hands!Not so sure if I'd run Windows on it though, especially not if it's one of those M$-Store locked-in variants. Linux or Android desktops seem much more attractive and compatible with what I'd use it for.
Dual-boot my Whiskey Lake with WinX and Fedora/Cinnamon.
HStewart - Wednesday, May 29, 2019 - link
Qualcomm is emulated - slow speed for real application, a better fit is Chrome than Windows.Xyler94 - Wednesday, May 29, 2019 - link
Windows on ARM is not emulated, it emulates X86 written applications. Any native ARM applications will work really good. just not many ARM based apps yet for Windows. They're probably coming though.sgeocla - Wednesday, May 29, 2019 - link
Looks like the 3rd article about Ice Lake and comparing it Ryzen 3700u and Vega 10.And still no mention in the article about how Intel used more expensive RAM to get 56% more bandwidth and on par performance.
Ryzen 3700u using RAM would get 33% more performance and walk all over the Ice Lake iGPU.
sgeocla - Wednesday, May 29, 2019 - link
> Ryzen 3700u using the same RAM would get 33% more performance and walk all over the Ice Lake iGPU.The Hardcard - Wednesday, May 29, 2019 - link
That’s on AMD though for not supporting faster RAM. Hopefully Zen 2 APUs nave expanded DDR support, including LP versions.sgeocla - Wednesday, May 29, 2019 - link
That excuses to some degree Intel on a technicality but certainly not reviewers.Reviewer should know better than to just take these great leaps of performance on face value from Intel without mentioning the discrepancy comes from RAM speed, not the improvements in architecture.
These sites should be doing tech journalism not masked payed advertising.
nevcairiel - Wednesday, May 29, 2019 - link
Well, since you can't really control the RAM speed in mobile devices, the source of the performance might be interesting, but for and end user also, ultimately, irrelevant, no?qap - Wednesday, May 29, 2019 - link
There is no need for "excuses".It is AMDs technical limitation in the very same way, that intel is for example limited by compute units count in current generation of iGPUs compared to AMD.
You can say "if intel had same ammount of compute units as amd, they would be faster and it is unfair that amd has more compute units!" but you can only blame intel that they did not put more of them. Same goes for lack of RAM support on AMD part.
Rudde - Thursday, May 30, 2019 - link
Ian seems busy, as it seems he didn't do any own analysing. He only repeated what was shown in the slides. I'm looking forward to the deep-dive he promised.TristanSDX - Wednesday, May 29, 2019 - link
18% more IPC. Patching security holes helped to regain 15% of lost perf. Remaining 3% come from march improvements.PLS aks Intel when 10nm desktop CPU will be released (if nay)
coschizza - Wednesday, May 29, 2019 - link
15% of lost perf ?????????????????????????? not existtipoo - Wednesday, May 29, 2019 - link
"If looking at the geometric mean for the tests run today, the Intel systems all saw about 16% lower performance out-of-the-box now with these default mitigations and obviously even lower if disabling Hyper Threading for maximum security. The two AMD systems tested saw a 3% performance hit with the default mitigations. While there are minor differences between the systems to consider, the mitigation impact is enough to draw the Core i7 8700K much closer to the Ryzen 7 2700X and the Core i9 7980XE to the Threadripper 2990WX."https://www.phoronix.com/scan.php?page=article&...
HStewart - Wednesday, May 29, 2019 - link
Ice Lake have security fixes in hardware so stop with patch BS.Xyler94 - Wednesday, May 29, 2019 - link
Read his comment carefully. He's saying due to the Hardware patches, they were able to regain 15% IPC lost to the mitigation. Keep in mind Intel is comparing vs Skylake, not Whiskey Lake, which if Intel is to be believed, Whiskey Lake had some hardware fixes for Meltdown and other vulnerabilities.So his comment isn't BS. It's a play on the fact the software/microcode fixes caused slowdowns. And because of the hardware fixes in Ice Lake, it regains that, plus 3%.
Targon - Friday, May 31, 2019 - link
The problem is that we don't know if Intel is comparing old Skylake performance pre-mitigations. Intel did say that not all mitigations were in place when that +18% IPC figure was done. Software/microcode fixes will still result in performance degradation, so honestly, at this point we can't really know how much of an improvement Intel has come up with. We DO know that since these are laptop chips, no consumers or review sites will be able to test them for at least three more months.Xyler94 - Friday, May 31, 2019 - link
Spectre and Meltdown are over a year old, I expect Intel would have had those fixes in place. ZombieLOAD and the rest of MDS I don't think so.RSAUser - Saturday, June 1, 2019 - link
That assumption fails as Intel can not do it and then boast higher gains with a nice Asterix.silverblue - Wednesday, May 29, 2019 - link
Ice Lake has hardware fixes for Zombieload?nonpracticingee - Thursday, May 30, 2019 - link
The IPC increase is due to the cache increases and it appears other improvements to the back end. It's similar to how Apple has been increasing performance for its in house chips over the past several years. Give it more cache, make it wider, watch it considerably outperform its predecessor.The IPC improvements are visible over Whiskey Lake which does have hardware mitigations. The geekbench scores show upwards of 20% IPC gains for single core but the frequencies of Ice Lake CPUs are a lot lower and thus the performance is roughly the same as Whiskey Lake.
sorten - Wednesday, May 29, 2019 - link
Perhaps not exciting for everyone, but I'm hopeful because of the TB3 integration. I'm curious to see how this impacts the design of the Surface Pro 7. Unfortunately we're probably waiting until Spring 2020 to find out.jeremyshaw - Wednesday, May 29, 2019 - link
It's MSFT. I'd expect them to screw it up and just not enable TB3 or Type-C at all.sorten - Wednesday, May 29, 2019 - link
Unfortunately a real possibility. MS has become frustratingly conservative with hardware.katsetus - Wednesday, May 29, 2019 - link
Wait. Hold on.18% IPC increase but vs 15W i7-8665U the turbo frequency is down 900 MHz and base frequency is down 600 MHz (in case of the 1065 shown in the table). Going by turbo frequencies:
3.9/4.8=0.8125
0.8125*1.18=0.95875
So it is slower than i7-8665U. Only at the maximum 4.1 GHz, will speed parity be achieved. Although, yes, I am comparing it to the processor with the highest frequencies and hence price.
schujj07 - Wednesday, May 29, 2019 - link
Even the 8th Gen i5's have higher boost than 4.1GHz. Across the board the 10th Gen has lower clocks than the 8th Gen. Sure IPC increases will help mitigate the lower clock speed, but it is alarming how much lower the clocks are on 10nm+ from Intel. These low clocks are probably why there isn't a desktop CPU being released since it won't clock high enough to overtake the 9th Gen in absolute performance.flgt - Wednesday, May 29, 2019 - link
Saw an interesting lecture the other day. I didn't realize that all things being equal (ie you don't introduce a new structure like FinFET) that transistor performance no longer scales with nodes. You go to a smaller node you get a slower transistor, since the voltage can no longer scale at the same rate. So maybe something brilliant comes along but it looks like peak clock frequencies will be coming down. The only reason to scale at this point is economics, and even that is starting to become questionable if the large investment now required for the next node cannot show returns.abufrejoval - Wednesday, May 29, 2019 - link
Exactly, and it's been like that for many years already.Actually the 4.8 GHz Turbos on current Whiskey Lakes are just downright crazy and the result of Intel engineers spending years on getting that type of performance via physical tuning out of an "old engine". Those speeds also require 25 Watts on my i7-8565U, so they aren't sustained beyond a couple of seconds.
Most of that physical tuning simply cannot be carried over to a new process node: With EUV and everything else, it's just a completely new ball game.
Density gains hopefully will pay off the additional fab cost and allow widening those CPU structures for the IPC gains, additional cores and beefier GPUs.
Speed and density used to be linked in the good old days up to perhaps 45nm, but they have become rather antagonistic.
I won't predict speeds won't climb to 5GHz again (evidently you just need to leave the engineers alone with a new process node for long enough), but it's going to be harder with every shrink.
edzieba - Wednesday, May 29, 2019 - link
Not just performance, but cost/transistor has been increasing per node since 28nm. Expecting die shrinks to give you cheaper faster chips has not been true for many years now.name99 - Wednesday, May 29, 2019 - link
No it has not! If that were true, what do you think would be the cost of, say, an A12X chip with 12 billion transistors?Designing for each smaller node has become more expensive, but that’s a very different claim.
Meteor2 - Thursday, June 6, 2019 - link
It's a hell of a lot. Have you seen the price of an iPhone X or iPad Pro?!Both design and manufacture have become much more expensive. That's not a secret.
name99 - Wednesday, May 29, 2019 - link
This is not a surprise. Why do you think Apple and ARM have been so aggressive about improved performance through increased IPC rather than frequency? Apple A12 is ~4x as fast as A7, but only twice the frequency. The other half of the performance jump is better IPC.Intel had access to exactly the same info. But they kept up the mad push for higher frequency, apparently learning nothing from the Pentium4...
Phynaz - Wednesday, May 29, 2019 - link
You’re an idiot.Shut up.
tipoo - Wednesday, May 29, 2019 - link
Is the stated IPC and per core improvement with the speculative exploit patches applied to older systems?nonpracticingee - Thursday, May 30, 2019 - link
If you look at the Geekbench scores comparing Whiskey Lake (which has some hardware mitigations) to Ice Lake, an IPC improvement of over 20% is visible over Whiskey Lake. However due to its lower clock rates, the single core performance is 2% lower and multicore 5% higher. Hard to market single digit performance increases when the selling point for Coffee Lake was something along the lines of "100% more cores and 60% more performance!" 5% is hard sell. GPU and platform improvements are easier to make a case for.Rudde - Thursday, May 30, 2019 - link
Using Geekbench for IPC comparision is a stretch. You need to look at the individual scores to be able to make any judgement about increases in IPC.CSMR - Wednesday, May 29, 2019 - link
> 32 EU parts are listed as "Ice Lake U UHD", while 48 and 64 EU parts are "Ice Lake U Iris Plus"> By increasing the number of execution units from 24 EUs to 64 EUs, Intel is promising over a 2x boost in graphics performance
This is a marketing gimmick then, comparing older non-Iris parts with new Iris parts. If they had compared older Iris parts with new non-Iris parts there would be a reduction in EUs from 48 to 32.
Also key is how much memory is going to be in the the Iris parts. Iris Pro 6200 had 128MB. If these are called "Iris Plus" then the memory could max out at 64MB.
jeremyshaw - Wednesday, May 29, 2019 - link
I'd venture it doesn't matter as much. ~60GT/s vs 100GT/s isn't nearly the advantage it used to be (especially considering memory architecture differences between DDR4 and Chrystalwell derivatives), if it even is an advantage at all.Irata - Wednesday, May 29, 2019 - link
I am wondering about TDP on these parts: At least for Desktop CPU, Intel's TDP is given for the base frequency. Does this apply for these mobile CPU as well ?This would be interesting seeing how e.g. the i7-1065 G7 has a base frequency of 1.3 Ghz, vs. a boost frequency of 3.9 Ghz.
(meaning - is the 9/15/28 W @ 1.3 Ghz ?)
vFunct - Wednesday, May 29, 2019 - link
This is will probably go into a MacBook redesign, perhaps with an OLED display. The MacBooks are due for a redesign later this year.tipoo - Wednesday, May 29, 2019 - link
That it was just refreshed makes the timing a bit strange. Apple is refreshing faster now, but not 3-4 months between a refresh and a major redesign fast. I think they did it because the 15" parts are further off, this would only go in the 13", and they won't have the 13" updated alone.Rash1 - Wednesday, May 29, 2019 - link
Which 10nm SKUs suitable for including in Macbook/Air/Pros. The 15" Macbook Pros already supporting 6 and 8 cores i7 so are there equivalent 10nm ICE processors in performance wise can take place inside current 15" Macbook Pros or we have to wait well in 2020 ?zodiacfml - Wednesday, May 29, 2019 - link
Finally, things are heating up! That i3 as a dual core is a downer though. They could have simply made it a quad core with lower boosts and no HT.abufrejoval - Wednesday, May 29, 2019 - link
Any vendor could use an i5, disable HT and lower clocks for lower energy, if there was a market for it. Don't think Intel makes any "core"-class dies that don't have the physical assets for HT.Sychonut - Wednesday, May 29, 2019 - link
WHAT? No 14++++? I'm so disappointed at Intel.HStewart - Wednesday, May 29, 2019 - link
Yes we have Zen++++ in less than 2 years .Korguz - Wednesday, May 29, 2019 - link
sorry to burst your bubble HStewart. but its tsmc.. not intel.. tsmc may not have the same issues as intel has for the last FOUR years with their node improvments...
Xyler94 - Thursday, May 30, 2019 - link
First of all, Zen2 is not a fab node, it's an architecture.Secondly, he's being sarcastic. I know it's hard for you to wrap your mind around that concept, but it's pure sarcasm.
Santoval - Wednesday, May 29, 2019 - link
"Users might be surprised about this 18% number, given that Intel has historically only provided low single digit boosts in recent generations."No, none of us will be surprised, because none of us will compare the IPC of Ice Lake to the IPC of a CPU series that was released in *2015* (not 2016). We will compare Ice Lake's IPC to that of its direct predecessor, which Intel is reluctant to do.
"The reason for those single digit boosts comes down mainly to small parts of the core microarchitecture being widened."
Ice Lake-U almost certainly has single digit extra IPC over Whiskey Lake-U as well. Intel compared their single thread performance (not IPC) only in a single graph, and they reported that Ice Lake-U has a ~4% higher performance than Whiskey Lake-U at 15W. Given their well-known love for skewed scores the actual number must be 2% at best.
The low clocks their 10nm+ node is able to reach has eaten away nearly all the IPC gains of Ice Lake over the direct predecessor CPUs, but given the small IPC gains that have gradually accumulated since Skylake I strongly doubt these gains are more than 8% (which, assuming the actual performance gain is 2%, would imply 6% lower sustained clocks) or 9% tops (for 7% lower sustained clocks).
"It will be interesting to get hold of a system and to actually test/compare with Skylake, especially with all the security patches applied."
Er, why on Earth would you want to compare Ice Lake to Skylake?! Did you actually mean the Skylake arch?
AntonErtl - Wednesday, May 29, 2019 - link
You compare Sunny Cove/Icelake to Skylake, because Skylake is the current microarchitecture. Skylake, Kaby Lake, Coffee Lake, Coffee Lake Refresh, and Whiskey Lake (did I forget any?) all have the same microarchitecture, and therefore the same IPC. There are some Meltdown/Spectre Mitigations in Coffee Lake Refresh AFAIK, but otherwise the microarchitecture is untouched.HStewart - Wednesday, May 29, 2019 - link
I think they did make some minor architexture improvements since Skylake related to instructions and such.alufan - Wednesday, May 29, 2019 - link
the time of intels reign is over AMD will be the leader for a few years again now, they have the same momentum intel had with the core 2 release and although intel has some comebacks i believe the advantage AMD has in multicore and design for heterogeneous computing will keep them ahead, have always bought AMD so am pleased that they are back on top again i dont hate intel i just prefer the actual response of amd desktops in the real world, the fact Ryzen is such a good chip just adds the cherry to the cake, awaiting 7nm release to order :)Phynaz - Wednesday, May 29, 2019 - link
Why would you post in an article about intel chips, talking about how big of an AMD fanbitch you are?Korguz - Wednesday, May 29, 2019 - link
Phynaz HStewart does it all the time, and hes and intel fan boy.. how come you arent saying the same thing to him in the AMD based articles ????Korguz - Wednesday, May 29, 2019 - link
hey HStewart learn how to spell.. its architeCture NOT architexture !!!! will you ever get it right ????paulemannsen - Thursday, May 30, 2019 - link
He is a bot. You are wasting your time. That ridiculous algorithm of his should have been banned long ago.Korguz - Friday, May 31, 2019 - link
not a bot.. just can only see intel, and sees intel as the god of the cpu... can do nothing wrong, never makes mistakes, is always right, etc...he has been asked a few times about the many wrong intel has done over the years.. and he either never replies.. or tries to twist, bend and make up facts to put intel back into the light...
StormRider45 - Wednesday, May 29, 2019 - link
An 18% jump in IPC against... Sky Lake which debut in 2015. This is disappointing and misleading to say the least. Not sure why Anandtech didn't point this out.Korguz - Wednesday, May 29, 2019 - link
why would they ?? there is HStewart for that :-) :-):-)tipoo - Monday, June 3, 2019 - link
Except when you realize IPC has been the same since then. It's clock speeds that have risen. The core architecture is largely the same since Skylake, so an 18% IPC bump over that is an ~18% IPC bump over everything since.outsideloop - Wednesday, May 29, 2019 - link
Maybe Asus/ASRock/Gigabyte will make a desktop board supporting Ice Lake mobile processors, like back in 2003, when the Pentium M mobile chips were introduced.The Yonah/Banias cores were faster/lower power/better than the Prescott Netburst P4 desktop chips that were starting fires at the time.
enzotiger - Wednesday, May 29, 2019 - link
compared to the original Skylake core released in 2016Really? 2016?
eastcoast_pete - Wednesday, May 29, 2019 - link
Ice Lake-U? What do you mean, Ice Lake-Me? Ice Lake-Yourself!Just couldn't help it.
eva02langley - Wednesday, May 29, 2019 - link
10nm barely match 14nm+... interesting...ballsystemlord - Wednesday, May 29, 2019 - link
1 grammar mistake:Intel's partners however have given us some specifications, and it shows in interesting adjustment in the naming scheme.
"an" not "in".
Intel's partners however have given us some specifications, and it shows an interesting adjustment in the naming scheme.
ballsystemlord - Wednesday, May 29, 2019 - link
1 grammar mistake:Intel's partners however have given us some specifications, and it shows in interesting adjustment in the naming scheme.
"an" not "in".
Intel's partners however have given us some specifications, and it shows an interesting adjustment in the naming scheme.
Ananke - Wednesday, May 29, 2019 - link
Intel has done this already - during the Pentium time, when AMD introduced multicore Athlon, Intel came with the completely new Core architecture in low wattage CPUs for mobile, and more than a year later they used Core in desktop and newer node process. Then, it was complicated because of agreements with Rambus, but essentially the process was the same.wr3zzz - Wednesday, May 29, 2019 - link
I am doubtful about the readiness of Y-series. XPS 13 2-in-1 has been the flagship Y-series device for years but Dell just added fans to its new Gen10 model.tipoo - Monday, June 3, 2019 - link
I wanted them to tbh, fans will bring it more in line with the performance of the non 2 in 1.zodiacfml - Wednesday, May 29, 2019 - link
Lukewarm as AMD has no 7nm for mobile for a while but glad to see nice upgrades with IGP.Oh, that i3 is terrible, could have been a quad core with lower boosts or disabled HT, leaving the dual core to a Pentium or Celeron brand.
Gunbuster - Friday, May 31, 2019 - link
Possible they have a lot of bad 4 cores that are getting binned down to 2? That's going to be a rough chip. Like original atom feel I'm thinking...pk1489 - Thursday, May 30, 2019 - link
According the roadmap of intel(include some information by Huawei leaking ),intel will unveils cooper lake-sp in 2019q4 and unveils ice lake-sp in 2020q1 for datacenter. Those production will sell in 2020Q1.nonpracticingee - Thursday, May 30, 2019 - link
I find it astounding that so many commenters in a site like this do not understand rudimentary concepts. Let’s clarify a few things:IPC = instructions per clock cycle (instructions divided by cycle)
Frequency/clock speed = Cycles per second, measured in Hertz (cycles divided by second)
When determining which is faster, you can measure either:
1) Instructions processed in the same amount of time (where a higher number of instructions is faster)
In this case, we can calculate the instructions through:
X instructions = instructions/cycle * cycles/second * seconds
2) The time it takes to process the same set of instructions (where a lower duration is faster)
X seconds = instructions / (instructions/cycle * cycles/second)
Of course, since the above are being measured per core, the other variable you can modify is the number of cores:
X instructions = (instructions/cycle * cycles/second * seconds)/core * y cores
Given the geek bench scores posted earlier, we have the following:
Single core Whiskey Lake i7-8665u, cTDP up to 2.1GHz, boost 4.8GHz - 5341
Single core ICL i7-1065G7 cTDP up to 1.5GHz, boost 3.9GHz - 5234
Multicore Whiskey Lake - 16485
Multicore ICL - 17330
Single core IPC based on base clocks: ICL IPC is 37% higher
Single core IPC based on boost clocks: ICL IPC is 21% higher
Raw single core performance: ICL is 2% slower
Raw multicore performance: ICL is 5% faster
I don’t have the boost clock based on active cores table for ICL, so we can’t accurately extrapolate the data there. But based on base clocks, there is a 5% performance increase with 30% lower frequencies, meaning there is a remote possibility that clock for clock it’s 50% faster in multicore scenarios.
In reality, the figures obtained are a combination of the chips running at different frequencies and for the short duration that the Geekbench tests run, the chips will primarily run close to the top rated boost clock speeds.
The IPC number is higher than what Intel is claiming because the work load that the CPU architects chose to model while designing the microarchitecture is different from what Geekbench is using. Based on the figures from the computed IPC based on boost clocks, they are similar though. The modifications they make to the microarchitecture are informed by the work loads they choose and there may be regressions in some work loads if they consider the improvements in other work loads to be more valuable.
Note: This only applies when comparing the same set of instructions. If the instruction set architecture (or even just the instructions used, like vectorized vs scalar instructions) is different, then the number of instructions processed is not equal to the amount of work done. E.g. if in one chip with one ISA, the same logical operation requires 1 instruction to perform that takes 6 cycles (IPC is 1/6), while another chip with a different ISA requires 3 instructions that take 2 cycles each (IPC is 1/2), the second chip has a higher IPC but in terms of actual work done, they are identical.
Both Intel and AMD implement the same ISA (with a few differing instructions) and thus can be directly compared for the most part. Recall that when AMD and Intel were competitive in the early 2000’s, Intel heavily marketed their higher frequencies and AMD named their Athlon XPs with names that didn’t represent the CPU frequencies because they had lower frequencies but higher IPC. AMD provided better value then, and with Ryzen 3, will appear to do the same again for the desktop market.
To generalize the Intel chips we’ve seen over the past few years we have:
Sandy Bridge
Ivy Bridge = Sandy Bridge + better GPU + 22nm
Haswell
Broadwell = Haswell + 14nm
Skylake
Kaby Lake = Skylake with higher frequencies
Coffee Lake = Skylake with higher frequencies, with more cores
Coffee Lake Refresh = Skylake with higher frequencies, with more cores, with more cores
What Intel was targeting to have is:
Sandy Bridge
Ivy Bridge = Sandy Bridge + better GPU + 22nm
Haswell
Broadwell = Haswell + 14nm
Skylake
Cannon Lake = Skylake + AVX512 + new memory module + 10nm
Ice Lake
In the amount of time between Skylake and Ice Lake, Intel could have finished development and begun shipment of yet another microarchitecture (beyond Ice Lake). However, because Intel’s microarchitectures are tightly coupled with a process node and Ice Lake was designed with the 10nm libraries, Intel could not produce chips with a new microarchitecture in 14nm. Since they are using the same microarchitecture and, thus, had constant IPC, the only remaining variables they could change to improve performance are: A) the frequency, and B) the number of cores. Kaby Lake modified variable A, Coffee Lake and Coffee Lake Refresh, modified A and B.
From the power usage perspective, we have the following:
1) shorter amount of time to perform the same work load, given higher frequencies and more cores
2) longer amount of time CPU spends idle thus longer amount of time in low power states
3) higher peak power usage because of more cores running at the same time
4) lower average power usage because CPU is idle longer
This plays a large part in Intel’s logic of keeping the TDP number identical. Obviously, if they defined TDP to represent peak power usage, it isn’t possible to keep the same figure if you have more cores running with mostly the same manufacturing node. If a core requires 2W to run, logically, running 2 cores will not also only require 2W at the same node.
Because the 10nm launch is so delayed and we’ve had Coffee Lake, et al to compensate, we have the following effect: die sizes are larger, but wafer sizes are the same. This means fewer chips per wafer, and thus lower capacities. Add to that binning for products like the 9900KS, and we have to go through multiple wafers to win the silicon lottery and get dies that can be placed in halo products. This in turn affects supplies to customers, where Intel wasn’t able to cope with the demand. This has caused Intel to invest in more 14nm fabs. This will likely cause Intel to produce more 14nm products for an even longer duration to make up for the costs of investing into more 14nm fabs.
Once Intel decided it would move away from the Tick-Tock strategy and embrace Process + Architecture + Optimization, they told us that 10nm would have lower performance than 14++ and 10+ would be roughly equivalent to 14++. This appears to be because 14+ and 14++ products allowed high frequencies that 10+ products are compensating for through higher IPCs. It is easy to see why Cannon Lake became a non-starter. The same IPC with lower clocks meant lower performance per core and low yields meant you had to keep the die size tiny (and consequently core count low) to have a viable die. Fewer cores, lower frequencies, and high costs means Cannon Lake was always going to be a disaster by the time we got Coffee Lake. Ice Lake with its higher IPC is able to compensate for the lower frequencies to provide comparable performance. The marketing for Ice Lake is touting Gen11 graphics and improvements to the platform because the CPU performance is essentially identical due to the less mature node.
TLDR
IPC gains are not hogwash. Intel is comparing ICL to SKL because Intel has ONLY been shipping SKL variants for 4 years now. In fact, IPC gains are the only reason Intel can afford to ship 10nm products now from a user perspective.
ICL will perform slightly slower for single threaded applications because of lower clocks but perform slightly faster for multithreaded workloads because IPC gains compensate for the lower clocks.
This doesn’t mean ICL itself underperforms. It’s currently hamstrung by the maturity of the node.
Hixbot - Thursday, May 30, 2019 - link
Wow, excellent post. I saw how much text and almost didn't read. but glad I did read it. Thanks.Rudde - Thursday, May 30, 2019 - link
Is it reasonable to count AES and AVX-512 to Ice-lakes advantage? Ice-lake has better AES acceleration (21%ST/ 33%MT improvement in Geekbench compared to i7-8665U). Ice-lake gets a boost in GEMM and FFT from AVX-512 (curiously it performed worse in ST GEMM than i7-8665U).My point is that support for specialised code improves the score of the Ice lake, making a general comparision harder. It is a nice boost in workloads that support it, but those are few and far between at the moment. It is also worth pointng that AVX instructions doesn't run at the same clock speed as integer instructions.
nonpracticingee - Thursday, May 30, 2019 - link
I wouldn't say they're unreasonable if they accelerate common workloads. Intel's been marketing how its specialized instructions improve performance since MMX. AMD's done the same thing with 3DNow! On that note, I don't think Intel's leaning heavily on AVX-512 for their IPC improvement claims. The product brief they released has a disclaimer about AVX lowering clock speeds and people have reported that AVX-512 is causing their CPUs to throttle, slowing down execution of even non-AVX code. If thermal issues are still a problem with AVX-512, they're probably amplified in Ice Lake because of the smaller die sizes. It's harder to dissipate heat when the surface area is smaller. Hopefully Ian can score some samples soon and we'll get to see real results. If anything, cache improvements and a better back end probably account for most of the IPC gains. We've seen what wonders that does for every generation of Apple A chips. Again, I'm eager to see what Ian has to share.zodiacfml - Thursday, May 30, 2019 - link
Yes. I guess we just have younger readers and don't understand what is going on with Intel. Aside from that, we haven't seen proper reviews yet and AMD will delay 7nm mobile. IceLake is important, though I don't like specs for the i3 chip with only two cores.Targon - Friday, May 31, 2019 - link
No delay from AMD on laptop chips, but AMD always starts with high end desktop to get the chips into systems quickly. If AMD is planning on Navi going into the 7nm laptop chips for the next generation, making sure that the GPU drivers work will be good for OEMs as well. The question is when AMD will release the new laptop chips. AMD released the 3000 series laptop chips(based on 12nm Zen+ and Vega) in January of 2019. If AMD gets the next generation out in November of 2019, that would be a positive thing.Targon - Friday, May 31, 2019 - link
A big factor that makes the stated numbers more of a challenge is security mitigations, and not having all mitigations in place when Intel stated the +18% IPC figure. This implies that the +18% figure isn't going to be accurate once the mitigations are in place. Most people are not saying that there are no improvements, but are pointing out that the actual IPC improvement may be completely offset by security mitigation numbers combined with lower chip frequencies. To make it even more difficult, Intel is known for setting TDP figures that only apply to the base frequency. If base frequency is only 1.1GHz for a 15W TDP chip, and that 15W is only accurate at the base frequency, then is Intel actually doing well compared to an AMD Ryzen 5 3500U or Ryzen 7 3700U in laptop performance?nonpracticingee - Friday, May 31, 2019 - link
You're right that _some_ of the comments were theorizing that the security mitigations account for most of 18% figure. I've previously responded to some of those comments to point out that the Geekbench numbers quoted are from Whiskey Lake and Ice Lake. _Both_ have some security mitigations in hardware. As I mentioned in my post, you'll see an IPC improvement higher than what Intel claimed because of a difference in the workload being tested. I did, however, also mention that there's barely any improvement visible when looking at the raw performance because of the low frequencies. This is why I said the IPC gains are the only reason Intel can afford to ship 10nm products. They can't sell something that performs worse but they also can't market single digit performance improvements and slight regressions so they focus on the GPU and platform improvements.The main reason I posted anything at all is because it's frustrating to see that people don't have a clue what IPC means. While _some_ of the posts brought up spectre/meltdown/mds, a lot of the others appear to think that each set of chips Intel has released over the past 4 years had IPC gains. I explained that Intel had to keep the IPC constant and rely on the two other factors they could change to improve performance: a) frequency, and b) core count.
In terms of laptop performance, it'll be hard to get a fair comparison regardless because OEMs tend to put AMD chips in cheap laptops with low end specs. Ice Lake does have a process advantage over the mobile Ryzen 3 chips since those are still manufactured at the 12nm node. We'll need to wait for reviews of actual products to see the difference. Moreover, the overall platform is a bigger deal in mobile products because of battery life concerns. In terms of laptop energy consumption, it isn't just the CPU that matters. Intel has integrated the functionality of more chips into its new PCH. This means that there will be fewer chips, lower overall power draw, smaller motherboard footprints, and lower costs for manufacturers. Integrating WiFi and Thunderbolt into the PCH alone would save a considerable amount of power.
tipoo - Monday, June 3, 2019 - link
Yup. I'm surprised so many in the AT audience don't understand this. IPC has been the same since Skylake, it's the clock speeds/boosts that have risen, so 18% over Skylake is very nearly 18% over everything since on Instructions Per Clock.MarjorieEHarness - Thursday, May 30, 2019 - link
you need a online job so please visit this site.......... online-3blucephone - Friday, May 31, 2019 - link
K'Pblucephone - Friday, May 31, 2019 - link
Here we will control on the most proficient method to introduce stock ROM on BLU Studio Mega 2018. The guide is straightforward and simple to pursue. You can watch our total video direct.https://blucellphones.us/download-blu-usb-drivers-...
weebcoup - Wednesday, June 5, 2019 - link
decide when and where to apply charges. Weebly gives and keeps up current city, state and region level assessment rates for the United States and Canada. For every single other nation, charge rates can be characterized by the vender.decide when and where to apply charges. Weebly gives and keeps up current city, state and region level assessment rates for the United States and Canada. For every single other nation, charge rates can be characterized by the vender.
decide when and where to apply charges. Weebly gives and keeps up current city, state and region level assessment rates for the United States and Canada. For every single other nation, charge rates can be characterized by the vender.
https://reviewsterr.com/weebly-coupon-code-discoun...
cobedrowlight - Saturday, June 8, 2019 - link
Driven Packaging Market by Package Type (SMD, COB, CSP), Packaging Material (Lead Frames, Substrates, Bonding Wire, Encapsulation Resins), Application (General Lighting, Automotive Lighting, Backlighting), and Geography - Global Forecast to 2023https://420reap.com/led-grow-lights/marshydro/
foodlve - Saturday, June 15, 2019 - link
"My grandma made jugs of this," says Tara, "and my sister and I, joined by our canine, would bounce from house to house in Belgaum gifting them to her companions. Consequently, they gave us their custom made summer specialities like Coorgi severe orange (kaipuli) squash, sharp mango pickle, or Goan dried fish pickle." https://pandareviewz.com/sindhi-food-cuisine/sleepwellfree - Saturday, June 29, 2019 - link
the administrators. By making sense of how to manage your time effectively, handle stress in a beneficial way, and keep up a calm, elevating angle, you'll have the alternative to rest better around night time. www.sleepnpeace.comquadibloc - Friday, July 12, 2019 - link
Ah, they aren't actually shipping these 10nm laptop chips just yet. However, this does mean that all those "leaks" of chips with designations like i7-10700F are fakes, since the naming scheme for the 10th generation is keeping things at four-digit numbers!basicshosting1 - Thursday, October 10, 2019 - link
A portion of the low, low costs you see on the greeting page for these web facilitating administrations are early on offers. At the point when the special time frame closes, the customary costs can be essentially higher. One essential special case is Complete Internet Solutions, which guarantees, "When you sign up your cost will never go up." <a href="https://wphostingbasics.com/bluehost-high-traffic-... traffic site hosting</a>