Qualcomm Windows on Snapdragon: New 7c & 8c SoCs for sub-$800 Laptops
by Anton Shilov on December 6, 2019 4:15 PM EST
Last year Qualcomm introduced its flagship Snapdragon 8cx platform for premium always-connected PCs (ACPCs) that packed the best technologies that the company had to offer at the time. Being a no-compromise solution, the Snapdragon 8cx was not meant for every ACPC out there, so this week the company expanded the lineup of its SoCs for laptops with the Snapdragon 7c for entry-level machines and the Snapdragon 8c for mainstream always-connected notebooks.
Qualcomm aimed its Snapdragon 8cx primarily at flagship devices ACPCs and therefore maxed out its performance and capabilities, as well as offering the ability to add a 5G modem inside. To day the SoC has won only three designs: the Lenovo 5G laptop (which is yet to ship), the Microsoft Surface Pro X (which uses a semi-custom version called SQ1), and the Samsung Galaxy Book S — all of which are going to cost well over $1000.
In a bid to address more affordable machines, Qualcomm will roll-out its slightly cheaper Snapdragon 8c SoC that is the same silicon as the 8cx, but will feature a tad lower performance. The 7c by comparison is a new chip that will also have a smartphone counterpart, and is aimed at sub-$400 devices, according to analyst Patrick Moorehead. Qualcomm even stated that the 7c is going to target Chromebook equivalents, if not ChromeOS itself.
| Qualcomm Snapdragon Flagship SoCs 2019-2020 | ||||
| SoC | Snapdragon 8cx | Snapdragon 8c | Snapdragon 7c | |
| CPU | 4x Kryo 495 Gold 4x Kryo 495 Silver Up to 2.84 GHz |
4x Kryo 490 Gold 4x Kryo 490 Silver Up to 2.45 GHz |
8x Kryo 468 Up to 2.40 GHz |
|
| GPU | Adreno 680 | Adreno 675 | Adreno 618 | |
| DSP / NPU | Hexagon 690 | Hexagon 690 | Hexagon ? | |
| AI Perf Combined | 7 TOPs | 6 TOPs | 5 TOPs | |
| Memory Controller |
8x 16-bit CH LPDDR4X-4266 63.58 GB/s |
4x 16-bit CH LPDDR4X-4266 31.79 GB/s |
2 x 16-bit CH LPDDR4-4266 15.90 GB/s |
|
| ISP/Camera | Dual 14-bit Spectra 390 ISP 1x 32MP or 2x 16MP |
14-bit Spectra 255 1x 32MP or 2x 16MP |
||
| Decode Encode |
4K120 10-bit H.265 720p480 HDR Support |
4Kp60 ? HDR Support |
||
| Wi-Fi | Wi-Fi 5 | Wi-Fi 6 | ||
| Integrated Modem | Snapdragon X24 LTE (Category 20) DL = 2000 Mbps 7x20MHz CA, 256-QAM, 4x4 UL = 316 Mbps 3x20MHz CA, 256-QAM |
Snapdragon X15 LTE (Category 15/13) DL: 800 Mbps 3x20MHz CA, 256-QAM, 4x4 UL: 150 Mbps 2x20MHz CA, 64-QAM |
||
| External Modem | Snapdragon X55
(LTE Category 24/22) (5G NR Sub-6 + mmWave) |
- | ||
| Mfc. Process | TSMC 7nm (N7) |
7nm | Samsung 8nm |
|
The 8c is the same chip as the 8cx, but clocked slightly lower. The 7c by contrast is built on Samsung’s 8nm process, and will mirror the specifications of a mid-range mobile chip in 2020. We were told that the 7c chip isn’t exactly ready yet, although other press were told that demos that were supposedly on 7c devices in our briefing were actually running 7c silicon.
The 8c, being an 8cx variant, can be paired with Qualcomm’s X55 modem to enable 5G connectivity, although it will be up to the OEM in order to determine if the device will have both Sub 6 GHz and mmWave support.
Devices featuring the 8c and 7c should come to market in 2020.
Alex Katouzian, senior vice president and general manager of mobile at Qualcomm Technologies, said the following:
“The mobile-first consumer wants an experience on par with a smartphone, and we have the innovation, the inventions and the technology to enable this experience for customers across price points.”
Related Reading
- Qualcomm Snapdragon Tech Summit Day 3 Live Blog: ACPC and XR
- Qualcomm Tech Summit, Day 3: Snapdragon 8cx, the New ACPC SoC
- Hands-On with Industry’s First 5G Laptop: A Lenovo with Qualcomm’s 8cx SoC and X55 Modem
- Samsung Announces Always-Connected Galaxy Book S Laptop with Snapdragon 8cx
Source: Qualcomm
Facebook
Twitter
RSS
53 Comments
View All Comments
thetrashcanisfull - Friday, December 6, 2019 - link
Any normal binary would have a regular x64 codepath anyway - otherwise it would break compatibility with the 90+% of x64 processors that don't implement AVX-512.peevee - Monday, December 9, 2019 - link
But then the performance of this code on all computational tasks will be abysmal.Wilco1 - Monday, December 9, 2019 - link
Well that means 99.99% of the PC market then since practically PC even supports AVX512...peevee - Tuesday, December 10, 2019 - link
I mean regular x64 codepath vs at least AVX2.Wilco1 - Tuesday, December 10, 2019 - link
AVX2 is not commonly available either (which Atom supports it?), the fallback would be an SSE(N) codepath, which the simulator does handle.In any case if you believe that vector performance matters here then you're quite wrong. The vast majority of x86 applications don't ever use vector instructions (being compiled with plain -O2 or even -Os settings).
SarahKerrigan - Friday, December 6, 2019 - link
No currently shipping consumer ARM has SVE. The overwhelming majority of currently shipping x86 lacks AVX512. Any sane x86 program has fallbacks to SSE4 or older, and Win/ARM64 presently offers binary translation for x86 but not x64 (which is rumored to come next year.)Performance of emulated x86 binaries is quick enough for programs that don't have a lot of heavy compute to be very usable. For those that are more processor intensive, the story is obviously different. Adobe is going to be doing a native ARM64 port of Creative Cloud, which may be the single largest pain point.
AlyxSharkBite - Saturday, December 7, 2019 - link
If you were coding for AVX-512 on anything outside of a benchmark, you don’t have a fallback. You coded AVX-512 for a specific scenario that it enhances greatly it’s in so few CPUs you did it for a reason. If you are just looking for an uptick in vectorized performance you code AVX2 with a fallback to AVX.Reflex - Saturday, December 7, 2019 - link
These really aren't the devices you'd be running AVX specific applications on. If that's a significant use case for you, there are much better options.nevcairiel - Sunday, December 8, 2019 - link
Any sane developer would create a fallback, if anything to validate the AVX code (because its far easier to write/validate non-SIMD code, and then compare the SIMD against that). And most will also have a AVX2 version of their AVX512 code, just to broaden the usability.Flunk - Sunday, December 8, 2019 - link
The vast majority of developers don't write in assembly and are at the mercy of the compiler author (often Microsoft, Intel or the GNU compiler authors). AVX512 instructions are very limited in applicable usage and as far as I know, you need to actively choose to use it.