Original Link: https://www.anandtech.com/show/11571/the-intel-ssd-545s-512gb-review-64layer-3d-tlc-nand-hits-retail
The Intel SSD 545s (512GB) Review: 64-Layer 3D TLC NAND Hits Retail
by Billy Tallis on June 27, 2017 6:00 AM ESTToday Intel is introducing their SSD 545s, the first product with their new 64-layer 3D NAND flash memory and, in a move that gives Intel a little bit of bragging rights, the first SSD on the market to use 64-layer 3D NAND from any manufacturer.
The Intel SSD 545s is a mainstream consumer SSD, which these days means it's using the SATA interface and TLC NAND flash. The 545s is the successor to last year's Intel SSD 540s, which was in many ways a filler product to cover up inconvenient gaps in Intel's SSD technology roadmap. When the 540s launched, Intel's first generation of 3D NAND was not quite ready, and Intel had no cost-competitive planar NAND of their own due to skipping the 16nm node at IMFT. This forced Intel to use 16nm TLC from SK Hynix in the 540s. Less unusual for Intel, the 540s also used a third-party SSD controller: Silicon Motion's SM2258. Silicon Motion's SSD controllers are seldom the fastest, but performance is usually decent and the cost is low. Intel's in-house SATA SSD controllers were enterprise-focused and not ready to compete in the new TLC-based consumer market.
The Intel SSD 545s continues Intel's close relationship with Silicon Motion by being one of the first SSDs to use the latest SM2259 controller. Since the SATA interface is now a dead-end technology, the SM2259 is a fairly minor update over the SM2258 controller used by last year's Intel SSD 540s. The only significant new feature enabled by the updated controller is hardware end-to-end data protection that includes ECC on the controller's SRAM and on the external DRAM. This will make the 545s more resilient against corruption of in-flight data, but it should not be mistaken for the power loss protection that is typically found on enterprise SSDs.
The flash memory used in the Intel 545s is Intel's second generation 3D TLC NAND, a 64-layer design with a floating gate memory cell. Intel did not use their first-generation 32-layer 3D NAND in a consumer SATA SSD, but the 32L 3D TLC is at the heart of Intel's SSD 600p, their first M.2 NVMe SSD and one of the most affordable consumer NVMe SSDs.
Similar to the strategy Micron used last year when introducing the Crucial MX300, the Intel 545s initially brings a new generation of 3D NAND to the market with just a single SKU. The 512GB 545s is available now on Newegg, with other capacities and the M.2 SATA versions to follow over the next few weeks. The full lineup will include capacities from 128GB to 2TB in both 2.5" and M.2 form factors.
Intel will be using their smaller 256Gb 64L TLC die for all capacities of the 545s, rather than adopting the 512Gb 64L TLC part for the larger models. The 512Gb die is not yet in volume production and Intel plans to have the full range of 545s models on the market before the 512Gb parts are available in volume. Once the 512Gb parts are available we can expect to seem them used in other product families to enable even higher drive capacities, but it is reassuring to see Intel choosing the performance advantages of smaller more numerous dies for the mainstream consumer product range.
Meanwhile, over the rest of this year, Intel plans to incorporate 64L 3D NAND into SSDs in every product segment. Most of those products are still under wraps, but the Pro 5450s and E 5100s are on the way as the OEM and embedded versions of the 545s.
Intel seems to be in a hurry to get this drive out the door so they can claim to be the first shipping SSDs with 64-layer 3D NAND. At Computex we saw Western Digital announce their first 64L 3D NAND SSDs due to be available in Q3, and Toshiba is already sampling the XG5 M.2 NVMe SSD to OEMs. Earlier this month, Samsung announced the start of volume production of their 64-layer 256Gb V-NAND. By launching with retail availability this week, Intel has narrowly secured first place bragging rights. (It seems Intel and Micron might have an agreement to take turns introducing new 3D NAND, given that Micron was first to ship the 32L 3D NAND last year with the Crucial MX300.)
The downside is that this is a rushed launch; I've had the drive in hand for less than five days as of publication time, and that time spanned a weekend. Intel's press briefing on this drive was a mere fifteen hours before the embargo lift, and the slides included some changed specifications relative to the product brief that was delivered with the drive last week. As with several of their recent SSD launches, Intel is only providing the one-page product brief and is withholding the full specifications document from the general public and the press, but this time it might genuinely be due to the latter document not being ready instead of motivated by the questionable IP security concerns Intel cited earlier this year.
Intel SSD 545s Specifications Comparison | |||
Model | 545s 512GB | 540s 480GB | |
Controller | Silicon Motion SM2259 | Silicon Motion SM2258 | |
NAND Flash | Intel 256Gb 64-layer 3D TLC | SK Hynix 16nm TLC | |
Sequential Read | 550 MB/s | 560 MB/s | |
Sequential Write | Burst | 500 MB/s | 480 MB/s |
Sustained | 475 MB/s | 40 MB/s | |
Random Read IOPS | 75k | 78k | |
Random Write IOPS | 90k | 85k | |
TCG Opal Encryption | No | No | |
Power Management | Slumber and DevSleep | Slumber and DevSleep | |
Form Factor | 2.5" 7mm (M.2 later this year) | 2.5" 7mm and M.2 2280 | |
Write Endurance | 288 TB (0.3 DWPD) | ||
Warranty | 5 years | 5 years | |
Launch MSRP | $179 | $174 |
The most significant performance improvement Intel cites for the 545s over the 540s is in sustained sequential transfers where writes exceed the size of the drive's SLC cache. In the briefing for the 545s Intel claimed the 480GB 540s would drop to 40MB/s while the 512GB 545s is capable of maintaining 475MB/s. The numbers given for the 540s are lower than what the full product specifications from last year list (125 MB/s). Without access to the comparable document for the 545s we can't entirely explain this discrepancy, but the most plausible reason is that Intel is no longer measuring sustained write speed restricted to an 8GB span of the drive and that they are now instead using a more sensible test where the drive is full or nearly so. Either way, the 545s should be able to perform much better after its SLC cache is full.
Externally, the 545s looks like a typical Intel SATA SSD with only minor design variations. Internally, the density of Intel's 3D NAND is readily apparent from the PCB that occupies less than half of the case and features only four NAND packages. With 256Gb (32GB) per die, this works out to four dies per package. Even the largest 2TB model should be able to use this PCB with sixteen dies per package and populating the empty pad for a second DRAM package. The Intel SSD 545s uses thermal pads on all four NAND packages and on the controller.
The 512GB Intel 545s debuts with a MSRP of $179. This is slightly higher than the launch MSRP of $174 for the 480GB Intel 540s, but on a price per GB basis the 545s is cheaper, and since its launch the MSRP of the 540s has been driven up to $189 by the onset of an industry-wide NAND flash shortage. In this narrow context the MSRP for the 545s may seem reasonable, but its true street price will need to be substantially lower. Intel's 600p NVMe SSD is currently only $175 on Newegg. Since the 600p outperforms any SATA SSD for typical real-world desktop use, the 545s needs to do better than 35¢/GB. The competition based on Micron's 32L 3D TLC includes the Crucial MX300 for around 30¢/GB, and the Samsung 850 EVO 500GB happens to be on sale on Newegg today for $165 (33¢/GB).
This launch comes at a bit of an awkward time for us. I've retired our aging 2015 SSD testbed and moved all the custom and homemade power measurement equipment over to a new system. Windows 8.1 is out and Windows 10 is in, and our IOmeter synthetic benchmarks are being replaced with Linux-based FIO tests that are more suited to modern TLC SSDs with SLC caches. For the past few weeks I've been focusing my efforts on validating the new testbed and test suite against NVMe SSDs, so the arrival at short notice of a new SATA SSD left me with no relevant comparison data. Given the time available, I chose to prioritize the benchmarks that are most relevant to real-world usage and to run a small selection of competing drives through those tests. This review will be updated with more benchmarks as the drives complete them, and the new SSD 2017 section of our Bench database will be going live soon and will be populated with results from the dozens of drives in our back catalog over the coming weeks.
For now, this review includes our three AnandTech Storage Bench (ATSB) workloads run on the new testbed, SYSmark 2014 SE and idle power management tests. The Intel SSD 545s is pitted against its predecessor the Intel SSD 540s, and most of the SATA SSDs with 3D NAND that have been on the market: Samsung's 850 EVO and 850 PRO, the Crucial MX300 and the ADATA Ultimate SU800.
AnandTech 2017 SSD Testbed | |
CPU | Intel Xeon E3 1240 v5 |
Motherboard | ASRock Fatal1ty E3V5 Performance Gaming/OC |
Chipset | Intel C232 |
Memory | 4x 8GB G.SKILL Ripjaws DDR4-2400 CL15 |
Graphics | AMD Radeon HD 5450, 1920x1200@60Hz |
OS | Windows 10 x64, version 1703 |
- Thanks to Intel for the Xeon E3 1240 v5 CPU
- Thanks to ASRock for the E3V5 Performance Gaming/OC
- Thanks to G.SKILL for the Ripjaws DDR4-2400 RAM
- Thanks to Corsair for the RM750 power supply, Carbide 200R case, and Hydro H60 CPU cooler
AnandTech Storage Bench - The Destroyer
The Destroyer is an extremely long test replicating the access patterns of very IO-intensive desktop usage. A detailed breakdown can be found in this article. Like real-world usage, the drives do get the occasional break that allows for some background garbage collection and flushing caches, but those idle times are limited to 25ms so that it doesn't take all week to run the test. These AnandTech Storage Bench (ATSB) tests do not involve running the actual applications that generated the workloads, so the scores are relatively insensitive to changes in CPU performance and RAM from our new testbed, but the jump to a newer version of Windows and the newer storage drivers can have an impact.
We quantify performance on this test by reporting the drive's average data throughput, the average latency of the I/O operations, and the total energy used by the drive over the course of the test.
The Intel SSD 545s performs surprisingly well on The Destroyer, with an average data rate that surpasses the Samsung 850 EVO and potentially makes the 545s the new fastest SATA SSD with 3D NAND. The 3D MLC-based Samsung 850 PRO is still out of reach, but the 545s is a big improvement over the 540s and the Crucial MX300.
The Intel 545s also ranks between the Samsung 850 PRO and 850 EVO in terms of average latency, with slightly more than half the latency of the Crucial MX300 and an even wider margin over the Intel 540s.
Separating the latency according to reads and writes, we see that the 545s is much closer to the Samsung 850 PRO than the 850 EVO for reads. The write latencies of the Samsung drives and the 545s are all much lower than the Crucial MX300 or Intel 540s.
In addition to delivering great performance for a SATA TLC SSD, the Intel 545s brings a huge improvement in power efficiency. The 545s uses 12% less energy than the Crucial MX300 or Samsung 850 EVO over the course of the test. Only a few SSDs have ever completed The Destroyer on such a small energy budget, and only one other TLC SSD has been this efficient.
AnandTech Storage Bench - Heavy
Our Heavy storage benchmark is proportionally more write-heavy than The Destroyer, but much shorter overall. The total writes in the Heavy test aren't enough to fill the drive, so performance never drops down to steady state. This test is far more representative of a power user's day to day usage, and is heavily influenced by the drive's peak performance. The Heavy workload test details can be found here. This test is run twice, once on a freshly erased drive and once after filling the drive with sequential writes.
The Intel SSD 545s shows substantial improvement over the 540s in average data rates for the Heavy test. The 545s is not quite able to compete against Samsung as was the case for The Destroyer, but it comes much closer than its predecessor.
The 545s also fares far better when the test is run on a full drive than the drives that use the previous generation 32L 3D TLC (ADATA's SU800 and Crucial's MX300). Both of those drives beat the 545s when the test is run on an empty drive, but suffer a huge performance loss when the drive is full.
The latency measurements for the Intel 545s on the Heavy test tell a similar story to the average data rates. Intel is now in the same league as the other 3D NAND SSD, though still in last place. When the test is run on a full drive, the Crucial and ADATA drives fall apart and average latency spikes to the 2–4ms range while the latency of the Intel 545s is unaffected and remains well below 1ms.
The read latency of the older Intel 540s is not too far behind the 3D NAND drives and it handles being full with no trouble, but its write latency in either case is almost twice as high as the 3D NAND drives. The biggest improvement of the Intel 545s comes from cutting that write latency. The read latency also improved enough to surpass the Crucial MX300.
The energy usage of the Intel 545s during the Heavy test has it essentially tied for second place with the Samsung 850 EVO, while the Crucial MX300 has the clear lead when the test is run on an empty drive. The MX300's advantage disappears when the test is run on a full drive and the SU800's energy usage almost doubles.
AnandTech Storage Bench - Light
Our Light storage test has relatively more sequential accesses and lower queue depths than The Destroyer or the Heavy test, and it's by far the shortest test overall. It's based largely on applications that aren't highly dependent on storage performance, so this is a test more of application launch times and file load times. This test can be seen as the sum of all the little delays in daily usage, but with the idle times trimmed to 25ms it takes less than half an hour to run. Details of the Light test can be found here. As with the ATSB Heavy test, this test is run with the drive both freshly erased and empty, and after filling the drive with sequential writes.
The Intel 545s delivers a much faster average data rate on the Light test than the 540s, and is even slightly faster than the Crucial MX300. It isn't quite up to the level of the Samsung drives, but it's reasonably close. The ADATA SU800 takes first place here, showing that it is optimized for high peak performance at the expense of very poor performance under sustained heavy workloads.
The average latency of the Intel 545s on the Light test is nothing special, but that's still an improvement over the 540s, or the MX300 when the test is run on a full drive.
All of the TLC SSDs suffer from significantly higher read latency when the Light test is run on a full drive rather than an empty drive. The MLC-based Samsung 850 PRO is only slightly affected, and the Intel 545s is less severely affected than the Crucial MX300 or the Intel 540s. The empty-drive write latency of the 545s is significantly better than the 540s but still slightly behind the other 3D NAND SSDs. When the test is run on a full drive, the write latency of the Crucial MX300 spikes and the 545s ends up tied with the SU800 and trailing only the Samsung drives.
The Light test is easy enough that the Crucial MX300 has the best power consumption whether the test is run on a full drive or an empty drive. The Intel 540s and 545s are essentially tied for second place, with the Samsung 850 EVO right behind. The Samsung 850 PRO is the only true outlier here: it generally sacrifices some power efficiency to deliver the best performance, but the Light test doesn't stress it enough for that to matter.
BAPCo SYSmark 2014 SE
BAPCo's SYSmark 2014 SE is an application-based benchmark that uses real-world applications to replay usage patterns of business users in the areas of office productivity, media creation and data/financial analysis. In addition, it also addresses the responsiveness aspect which deals with user experience as related to application and file launches, multi-tasking etc. Scores are calibrated against a reference system that is defined to score 1000 in each of the scenarios. A score of, say, 2000, would imply that the system under test is twice as fast as the reference system.
SYSmark scores are based on total application response time as seen by the user, including not only storage latency but time spent by the processor. This means there's a limit to how much a storage improvement could possibly increase scores, because the SSD is only in use for a small fraction of the total test duration. This is a significant difference from our ATSB tests where only the storage portion of the workload is replicated and disk idle times are cut short to a maximum of 25ms.
AnandTech SYSmark SSD Testbed | |
CPU | Intel Core i5-7400 |
Motherboard | ASUS B250-PLUS |
Chipset | Intel B250 |
Memory | 2x 8GB Kingston DDR4-2400 CL17 |
Case | In Win C583 |
Power Supply | Cooler Master G550M |
OS | Windows 10 64-bit, version 1703 |
Our SSD testing with SYSmark uses a different test system than the rest of our SSD tests. This machine is set up to measure total system power consumption rather than just the drive's power.
The SYSmark performance scores for data analysis and media creation are relatively insensitive to storage performance, as shown by our test system's score with a mechanical hard drive exceeding the normalized score of 1000 for the SYSmark reference system that uses a SSD and coming relatively close to the performance of our SSDs. The differences between these SATA SSDs are all within the variation between runs.
The office productivity test also shows very little dependence on storage performance, but the gap between the SSDs and the hard drive is a bit wider than for the first two usage scenarios, and the hard drive's detrimental impact on performance was enough to offset the CPU and RAM advantages our testbed has relative to the SYSmark 2014 SE reference system.
The SYSmark responsiveness benchmark is the most sensitive to storage performance, and all of the SATA SSDs are about twice as fast as the mechanical hard drive. The difference between SSDs is just barely larger than the variation between test runs, but it's enough that the slowest of the three runs for the 545s was faster than the fastest of the three runs for the Intel 540s.
Energy Usage
The SYSmark energy usage scores measure total system power consumption, excluding the display. Our SYSmark test system idles at around 26 W and peaks at over 60 W measured at the wall during the benchmark run. SATA SSDs seldom exceed 5 W and idle at a fraction of a watt, and the SSDs spend most of the test idle. This means the energy usage scores will inevitably be very close. A typical notebook system will tend to be better optimized for power efficiency than this desktop system, so the SSD would account for a much larger portion of the total and the score difference between SSDs would be more noticeable.
Rather than judging these scores by their percentage difference, it's probably more useful to consider the raw number of Watt-hours saved by using a more efficient SSD. The difference here between the best and worst SSD is just under 1 Wh, compared to a typical ultrabook battery capacity of around 50 Wh. Since the overall performance scores with each SSD were so close together, we know that the test runs took essentially the same amount of time and the differences in energy usage are due mostly to differences in the idle power consumption of the SSDs.
Idle Power Consumption
Since the ATSB tests based on real-world usage cut idle times short to 25ms, their power consumption scores paint an inaccurate picture of the relative suitability of drives for mobile use. During real-world client use, a solid state drive will spend far more time idle than actively processing commands. Our SYSmark 2014 SE test covers this case, but because our testbed is a desktop system the SSD's contribution to the total power usage is small and hard to isolate from ordinary variation between test runs.
Our testbed doesn't support the deepest DevSlp power saving mode that SATA drives can implement, but we can measure the power usage in the intermediate slumber state where both the host and device ends of the SATA link enter a low-power state and the drive is free to engage its internal power savings measures.
We also report the drive's idle power consumption while the SATA link is active and not in any power saving state. Drives are required to be able to wake from the slumber state in under 10 milliseconds, but that still leaves plenty of room for them to add latency to a burst of I/O. Because of this, many desktops default to either not using SATA Aggressive Link Power Management (ALPM) at all or to only enable it partially without making use of the device-initiated power management (DIPM) capability. Additionally, SATA Hot-Swap is incompatible with the use of DIPM, so our SSD testbed usually has DIPM turned off during performance testing.
With SATA device-initiated link power management enabled, typical idle power consumption drops by a factor of ten, straining the resolution of our power meter. All of these SATA SSDs have very effective low-power idle states, but the 545s does show clear improvement over the 540s and Intel is now quite close to the Crucial MX300 and Samsung 850 EVO.
The Samsung drives have the best active idle power levels of this bunch, and the Crucial MX300 is not far behind. The Silicon Motion-based Intel 540s, 545s and ADATA SU800 are worst, though only the SU800 is trailing by a large margin. The 545s does draw more power than the 540s, which is not what we would expect from an updated controller design.
Idle Wake-Up Latency
Idle power consumption is not the only important metric of a SSD's power management capabilities. Devices cannot transition in and out of low-power states instantly. It takes time for components to switch back to higher clock speeds and resynchronize their data links to the rest of the system. At the extreme end, the several seconds it takes a hard drive to spin up obviously has a major impact on system responsiveness, and consequently operating systems have to be very conservative about when to put a hard drive to sleep, typically requiring at least several minutes of idle time during which the hard drive is wasting power.
The idle power transition time of solid state drives exists on a very different time scale, but it still matters to systems that are trying to minimize power consumption without sacrificing too much performance. When a SSD only takes a few milliseconds to enter and leave a power saving state, those power saving states can be used far more often, even during sustained interactive use like gaming. The cost of this approach is that far more I/O operations will incur the wake-up latency penalty.
Our equipment is not able to measure how long it takes a SSD to enter a low-power state; this would require very precise synchronization between the CPU and a high-resolution power meter. Timing how long a SSD takes to leave a low-power state is simple: perform one read operation every few seconds, and compare the latency with and without power management enabled. The difference, reported below, includes time taken to reestablish the SATA link and for the SSD to wake up, but does not include the time spent actively reading data from the flash memory since that will be the same whether or not the SSD had just been in a low-power state. Thus, this test is mostly a measure of the SSD controller and its firmware.
The Crucial MX300 has pretty good power consumption both at idle and under load, but its Marvell controller takes far longer to wake up from idle than its competition. At about 3.4ms, the MX300 is almost five times slower to wake up than the ADATA SU800, which has higher idle power consumption. The Intel 545s is a bit on the slow side at 1.4ms, about twice the wake-up latency of the ADATA SU800 or the Intel 540s. The total read latency for the 545s with power management enabled is just over 2.2ms.
Looking Forward
Our first look at the Intel SSD 545s with Intel's new 64-layer 3D TLC NAND flash memory has been very promising. Compared to its predecessor based on 16nm planar TLC NAND, the 545s is a big leap forward in performance. Where the Intel 540s was really only suitable as an entry-level consumer SSD, the 545s is a much better all-around performer.
The launch last year of the first generation of Intel/Micron 3D NAND did little to challenge Samsung's dominance at the high end of the SSD market. The onset last fall of an industry-wide NAND flash memory shortage drove prices higher and put many new products on hold, but a few models based on the Intel/Micron 32-layer 3D TLC were able to establish solid positions as great value options.
The Intel 545s hints that this generation of 3D NAND may have a much bigger impact on the market. Even with the limited testing we had time for, it is clear that the Intel 545s does not suffer from the performance pitfalls that mar the Crucial MX300 and ADATA SU800, both based on Micron's 32L 3D TLC. On some tests, the Intel 545s appears to be the first serious challenger to the Samsung 850 EVO's combination of high performance and good power efficiency in a TLC SATA SSD. The surprisingly strong performance of the Intel 545s on The Destroyer shows that Intel's new 3D TLC NAND can handle heavy workloads, while the drives based on the previous generation of Intel/Micron 3D TLC tend to only be competitive on lighter workloads. We eagerly await the announcement of a NVMe SSD based on this 3D NAND.
The Intel 545s also challenges some of our assumptions about Silicon Motion's controller and firmware. Silicon Motion was one of the best options for mainstream SSDs back when the market was dominated by MLC NAND, but their first two generations of solutions for TLC NAND sacrificed performance and power efficiency in a race to the bottom. With planar NAND on the way out and with a new iteration of 3D NAND and the controller and firmware solutions to support it, this may be the time for Silicon Motion to reestablish their reputation for great power efficiency and good mainstream performance at highly competitive prices.
However, it's too soon to determine whether the Intel 545s will actually be competing in the right price range. Based on the testing so far, the Intel 545s should be targeting a price near the Crucial MX300, which has spent most of the past year as one of the cheapest SATA SSDs available despite offering performance a step up from the entry-level planar TLC SSDs. The MSRP of $179 for the 512GB model is significantly higher than that target, but a fair price comparison will have to wait a few weeks for broader availability of the Intel 545s.
Intel has successfully launched a new generation of flash memory and a new mainstream consumer SSD. They haven't completely upended the SSD market like some of their SSD launches in the past and the impact will be muted at first due to limited supply, but this is still clearly a step forward. The bar has been raised a bit higher for the upcoming 3D NAND SSDs from Toshiba and Western Digital, and by the end of the year Samsung's SSD division should be feeling a lot more pressure than they have in a long time. Even if the NAND shortage will be keeping prices elevated into 2018, the market is moving forward in ways that will benefit consumers.