Qualcomm has Snapdragon 8cx beating Core i5-8250U in PCMark 10

[moinmoin]

Forget emulation, just forget it. Its is a stop-gap solution at best. The best dynamic translators i am aware of are still factor 2-3 slower than native. If we assume that ARM is 2x as effcient as x86, then with emulation you throw this advantage right out of the window ...period.
That does not mean, that applications are necessarily too slow for usage under emulation ...it is just that the advantage of ARM is gone.

When buying an ARM device you want to have highest performance in a power constrained device. Thats precisely what you get - when running native apps.

The Office benchmark numbers in OP were done using emulation and still beat Core i5 8250u according to Qualcomm. Yes, it's a stop-gap solution, but always better than no software support at all.

 

[soresu]

Thought I would drop in an update on mobile/embedded open GPU drivers, found this PDF link on Phoronix for those interested.

 

[Lodix]

Do we have an actual mm2 number for SD 855?

My bet would be it not being enormous. Pricing being bad is QCOM's thing.

It is 73.27mm^2.

Enviado desde mi SM-G975F mediante Tapatalk

 

[soresu]

The Office benchmark numbers in OP were done using emulation and still beat Core i5 8250u according to Qualcomm. Yes, it's a stop-gap solution, but always better than no software support at all.

This x10.

It's all well and good saying we need native, but a vast library of legacy x86 software is not something to sniff at.

Many apps will never get updates even for modern x86, let alone ARM native support.

With Google ANGLE pushing for a Desktop GL front end for their new Vulkan back end, hopefully WARM will get Vulkan support to allow them to gain significant GL support without much expended effort on their part (though it should be up to GPU driver makers to decide IMHO).

Do we have any idea what DX level support WARM has?

Obviously some degree of DX12, and likely DX11 - but what about DX10/10.1 and DX9?

 

[soresu]

It is 73.27mm^2.

Enviado desde mi SM-G975F mediante Tapatalk

Damn that is small, just slightly larger than Zen 2 CCX chiplet, albeit with all the extra stuff too (GPU, mem controller, modem, audio, video codecs, IPU, NPU/Hexagon DSP and whatever else is in there).

 

[AkulaMD]

As posted by @Gideon at https://forums.anandtech.com/threads/speculation-ryzen-4000-series-zen-3.2567589/post-39902153
Thought this is worth a dedicated thread as this market is one Intel is currently strongly focusing on.

Caveat: private event, no independent benchmarks outside of that so far.
https://www.windowscentral.com/snapdragon-8cx-benchmarks

"Interestingly, Microsoft Office is not natively compiled for ARM and technically runs emulated, meaning the Snapdragon 8cx was actually at a disadvantage."

"That said, the Snapdragon laptop was running at full HD while the Intel one was 2K, which is a bit unfair though it would not explain the nearly 7-hour difference either."

Youtube playlist

Very encouraging results. Hope the cheaper 7cx would be almost as good to enable mainstream adoption.

 

[soresu]

"That said, the Snapdragon laptop was running at full HD while the Intel one was 2K, which is a bit unfair though it would not explain the nearly 7-hour difference either."

TSMC 7nm explains a significant part of it at least.

 

[DrMrLordX]

Thought I would drop in an update on mobile/embedded open GPU drivers, found this PDF link on Phoronix for those interested.

That's at least interesting in its ability to tell you who cooperates with open source GPU driver efforts and who doesn't.

 

[Nothingness]

The Office benchmark numbers in OP were done using emulation and still beat Core i5 8250u according to Qualcomm. Yes, it's a stop-gap solution, but always better than no software support at all.

Are we sure Office is being emulated?
From https://www.notebookcheck.net/Windo...e-now-and-what-the-future-holds.413796.0.html

When we reviewed the Lenovo Yoga C630 ($850 MSRP), we found that software with ARM-native versions performed well. These included key software like Microsoft Edge and Microsoft Office, covering the vital combination of browser and office suite.

On the other hand, from https://www.techrepublic.com/article/windows-10-on-arm-what-we-learned-at-build-2018/

Some applications — like Microsoft Office — can mix x86 and Arm code using CHPE (Compiled Hybrid Portable Executable) DLLs with Arm64 code but 32-bit x86 interfaces; as a result, hundreds of third-party Office add-ins that expect Office to be an x86 application will work with Office on Arm too.

I have taken a look at x86->ARM translated code. It's quite nice, but the price to pay still is high (about 2x), so I doubt you'd get such good results by being fully translated on the fly.

 

[Thala]

This x10.
It's all well and good saying we need native, but a vast library of legacy x86 software is not something to sniff at.
Many apps will never get updates even for modern x86, let alone ARM native support.

Please keep references to my post in context. I was only refering to x64 emulation, which i even consider detrimental to the platform.

Do we have any idea what DX level support WARM has? Obviously some degree of DX12, and likely DX11 - but what about DX10/10.1 and DX9?

Under x86 emulation even older versions of DX are supported. You can for instance run Unreal (the original game from 1998) with DX driver. In addition you have OpenGL 2.0 (Microsoft Implementation, non accelerated).

Within the native environment you have DX9-12, however DX9 with some restrictions as the d3dx9.dll is not available.

 

[Thala]

I have taken a look at x86->ARM translated code. It's quite nice, but the price to pay still is high (about 2x), so I doubt you'd get such good results by being fully translated on the fly.

It can be even 3x and slower - still very fast for a dynamic translator. Regarding MS Office its looking much better as its using CHPE dlls and generally spend much execution time native.
It is looking not as favourable if the majority of execution time is spend in emulated code - like when running Povray and such.

 

[Shivansps]

Forget emulation, just forget it. Its is a stop-gap solution at best. The best dynamic translators i am aware of are still factor 2-3 slower than native. If we assume that ARM is 2x as effcient as x86, then with emulation you throw this advantage right out of the window ...period.
That does not mean, that applications are necessarily too slow for usage under emulation ...it is just that the advantage of ARM is gone.

When buying an ARM device you want to have highest performance in a power constrained device. Thats precisely what you get - when running native apps.

This is very simple, one thing is Chromebooks, another very diferent thing is the Notebook and Desktop market, that is dominated by Windows and Windows apps. You cant win those markets with power efficency alone, you need a good balance of performance, user experience, app support and price. And dont put your money on the very hated UWP.

So ARM for the highest performance in a power constrained device? Sure, still depends of what you are talking about, for something like a RPI 4, ARM is perfect, but in the notebook and desktop market that is not enoght. ARM first need to take the Chromebook market and x86 still dominates it, and i guess there is a reason for that, price-performance is the reason. Why they are still using Braswell? It has a higher performance than ARM chips for the same price, thats the ONLY reason for using x86 in that market.

RPI 4 A72 cores provides about the same performance than a 5 year-old Sempron 3850 or a Atom Z3735F whiout emulation. Thinking about this, you probably need a quad core A77 to have the same perf under emulation.

 

[soresu]

thats the ONLY reason for using x86 in that market.

I suspect its more likely a long standing financial incentive program (contra revenue) from Intel to keep Chromebooks from becoming another entrenched beach head of ARM, as Android already was by the time Intel finally got there.

With the support for Android apps in Chrome, given the prevalence of ARM code in Android apps, we should have seen far more ARM based Chromebooks, it's counter intuitive that the opposite is true.

The only thing stopping me taking the plunge for one is a high end ARM chip to succeed the aging MT8173 and RK3399 based designs (yes 3-4 year old SoC's are the best you can get currently for ARM chromebooks).

I would prefer SD 855, but I would settle for RK3588 if it is significantly cheaper.

 

[Shivansps]

I suspect its more likely a long standing financial incentive program (contra revenue) from Intel to keep Chromebooks from becoming another entrenched beach head of ARM, as Android already was by the time Intel finally got there.

With the support for Android apps in Chrome, given the prevalence of ARM code in Android apps, we should have seen far more ARM based Chromebooks, it's counter intuitive that the opposite is true.

The only thing stopping me taking the plunge for one is a high end ARM chip to succeed the aging MT8173 and RK3399 based designs (yes 3-4 year old SoC's are the best you can get currently for ARM chromebooks).

I would prefer SD 855, but I would settle for RK3588 if it is significantly cheaper.

AMD also has some designs wins with a ancient core whiout even trying. There is more than just contra revenue going on here. Im going back to the RPI 4... that cheap A72 quad core CPU has similar scores to a old Z3735F, thats a 1.3Ghz Quad Core Bay Trail... this most likely means that MT8173 and RK3399 are slower than a modern quad core Atom. Thats no good in my book, you probably need newer cores than A72 on the low end to match Atom price-performance, and price may not be that good, and this is were power efficiency goes out the window. This is problably why Intel small cores are doing that well in Chromebooks and why ARM has no penetration on notebook market. But this is going to change very very soon. But Intel also has Little.Big coming to fight back. Not sure if it going to be enoght, ARM premium chips already showed they can at least match non-emulation X86 performance.

 

[jpiniero]

Intel isn't doing contra-revenue anymore but you can be sure they aren't making much if anything on Chromebook processors.

AMD did manage to get a design win for Chromebooks but I imagine it had more to do with Intel's shortages than anything else.

 

[soresu]

Giving large discounts to maintain market share is barely any different from contra revenue - though once Snapdragon breaks in, it will become a free for all, Qualcomm is no weakling in the market compared to Rockchip.

Intel has the benefit of the standardised x86 platform though, if ARM ODM's get their act together and standardise they will stand a much better chance of competing.

 

[ElFenix]

Intel isn't doing contra-revenue anymore but you can be sure they aren't making much if anything on Chromebook processors.

AMD did manage to get a design win for Chromebooks but I imagine it had more to do with Intel's shortages than anything else.

i think AMD is just dumping garbage they have to buy from GF. probably darn near giving those processors away.

 

[Thala]

AMD also has some designs wins with a ancient core whiout even trying. There is more than just contra revenue going on here. Im going back to the RPI 4... that cheap A72 quad core CPU has similar scores to a old Z3735F, thats a 1.3Ghz Quad Core Bay Trail... this most likely means that MT8173 and RK3399 are slower than a modern quad core Atom. Thats no good in my book, you probably need newer cores than A72 on the low end to match Atom price-performance, and price may not be that good, and this is were power efficiency goes out the window. This is problably why Intel small cores are doing that well in Chromebooks and why ARM has no penetration on notebook market. But this is going to change very very soon. But Intel also has Little.Big coming to fight back. Not sure if it going to be enoght, ARM premium chips already showed they can at least match non-emulation X86 performance.

Not sure where you got your comparision from but they are way off. A Cortex A72 is way faster per clock than any BaytrailAtom or Silvermont/Airmont/Goldmont microarchitecture in general.
It is Goldmont+/Gemini Lake platform, which (slightly) outperfporms A72.

The fastest Gemini Lake implementation is Pentium Silver J5005 with a single core Geekbench Score of roughly 2200 @ 2.8GHz and 10W TDP. Thats roughly what we get for an 2.8GHz upscaled A72 as well.
In contrast if we upscale Z3735F from 1.8GHz -> 2.8GHz we ending up with a score of 15xx.

 

[Shivansps]

Im looking at results in Geekbeench 4... the RPI 4 has a little higher ST score but slower MT score than a Z3735F and thats the SLOWER Bay Trail Quad Core CPU and the one i have in my 5 year old $99 tablet.

With Active-Cooling the RPI 4 has higher ST and MT scores. But Again thats the SLOWER Quad core BT CPU, the Z3775 has a far higher scores than the RPI 4 even with active cooling. And thats still not the faster BT.

RPI4 passive
https://browser.geekbench.com/v4/cpu/13701853

RPI4 Active cooling
https://browser.geekbench.com/v4/cpu/13864995

Z3735F
https://browser.geekbench.com/processors/1642

Z3775
https://browser.geekbench.com/processors/1647

With 64 bit Linux those scores are going to be better, so i consider RPI4 A72 cores to be about equal to a quad core BT.

-this is also the reason of why i think RPI4 can run Windows 10 perfectly, if a GPU driver is developed-

 

[burninatortech4]

i think AMD is just dumping garbage they have to buy from GF. probably darn near giving those processors away.

Those chips are Stoney Ridge. Definitely giving them away at this point. It's 2019 haha.

 

[soresu]

Im looking at results in Geekbeench 4... the RPI 4 has a little higher ST score but slower MT score than a Z3735F and thats the SLOWER Bay Trail Quad Core CPU and the one i have in my 5 year old $99 tablet.

With Active-Cooling the RPI 4 has higher ST and MT scores. But Again thats the SLOWER Quad core BT CPU, the Z3775 has a far higher scores than the RPI 4 even with active cooling. And thats still not the faster BT.

RPI4 passive
https://browser.geekbench.com/v4/cpu/13701853

RPI4 Active cooling
https://browser.geekbench.com/v4/cpu/13864995

Z3735F
https://browser.geekbench.com/processors/1642

Z3775
https://browser.geekbench.com/processors/1647

With 64 bit Linux those scores are going to be better, so i consider RPI4 A72 cores to be about equal to a quad core BT.

-this is also the reason of why i think RPI4 can run Windows 10 perfectly, if a GPU driver is developed-

RPi4 uses 28nm, it's neither A72's ideal process nor a fair comparison point to a 22nm finfet or 14nm finfet atom where clock speed is concerned.

RPi is also not really manufactured for pure performance, you can get RK3399 boards that make paste out of it on Geekbench:

Link here.

 

[Nothingness]

Here is a comparison of Qualcomm SD845 (Cortex-A75) vs Atom J5005 (Gemini Lake): https://browser.geekbench.com/v4/cpu/compare/14465539?baseline=10520513

Both are from end of 2017/beginning of 2018. They are very close. Well except if you forget that the J5005 TDP is 10W.

 

[Thala]

Im looking at results in Geekbeench 4... the RPI 4 has a little higher ST score but slower MT score than a Z3735F and thats the SLOWER Bay Trail Quad Core CPU and the one i have in my 5 year old $99 tablet.

With Active-Cooling the RPI 4 has higher ST and MT scores. But Again thats the SLOWER Quad core BT CPU, the Z3775 has a far higher scores than the RPI 4 even with active cooling. And thats still not the faster BT.

At this point we can start to speculate, why the RP4 scores are that low - in any case they are not remotely representative for Cortex A72 performance. Just look at the RK3399 numbers linked by Soresu above - these are 1.6GHz numbers.
If you then further clean-up by clockspeed the distance gets even larger.

 

[DrMrLordX]

I'm still really looking forward to Rockchip's A76 SoC. All those 8cx machines look like "5G always-on" machines that don't even have wifi built-in. What a crock.

 

[soresu]

I'm still really looking forward to Rockchip's A76 SoC. All those 8cx machines look like "5G always-on" machines that don't even have wifi built-in. What a crock.

No WiFi built in?

That would be insane, no cell coverage is perfect, even in towns and cities buildings can interfere with the signal - I used to live on a street 10-15 minutes from the city centre of Swansea, and it had terrible reception for 4G.