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

moinmoin

Senior member
Jun 1, 2017
987
755
106
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
 
Last edited:
  • Like
Reactions: amd6502

naukkis

Senior member
Jun 5, 2002
232
79
101
8cx uses Cortex-a76? Pretty competitive already, and A77 brings another 20% of performance......
 

Arkaign

Lifer
Oct 27, 2006
19,795
325
126
If ancient Intel 14nm is this close, AMD 7nm Zen2 and beyond should absolutely annihilate this segment, just a matter of getting the SKUs out the door.
 

CluelessOne

Member
Jun 19, 2015
41
21
81
Was PC Mark running Office at 32 or 64 bit? I thought it needs to be 64 bit for it to be emulated. This is running in the S mode? UWP only?
 

Roland00Address

Golden Member
Dec 17, 2008
1,855
35
91
Why use a 2k display and what resolution do they mean by 2k? There are several different resolution numbers that could be used for the 2k number.
 

Roland00Address

Golden Member
Dec 17, 2008
1,855
35
91
According to CNX Software it's 2048x1080.
What laptop is that? There is a laptop in the video but I do not recognize it. Also why choose 2048x1080 as a panel makes no sense to me in my head, that is wider than 1920x1080 and that makes no sense as laptop for often laptops do not want to be wider, and why 2048? It seems a weird number to settle on for a custom laptop panel.
 

DrMrLordX

Lifer
Apr 27, 2000
13,015
2,688
136
If ancient Intel 14nm is this close, AMD 7nm Zen2 and beyond should absolutely annihilate this segment, just a matter of getting the SKUs out the door.
8cx vs Renoir? I like it. Or maybe the next generation beyond 8cx if Qualcomm is quick enough to update their platform.

Regardless, this thing is much faster than the 855. The ARMy is finally here, folks. Things could get . . . interesting.

Does anyone know if 8cx supports SVE at all?
 

Gnyueh

Junior Member
Feb 10, 2019
11
2
41
Was PC Mark running Office at 32 or 64 bit? I thought it needs to be 64 bit for it to be emulated. This is running in the S mode? UWP only?
32bit
MS cannot emulate x86-64 on ARMv8 for now.
So you cannot run x86-64 apps on 8cx
 

VirtualLarry

Lifer
Aug 25, 2001
44,749
3,812
126
Does the OS version disparity mean anything? I thought that I had heard that 1903 had a new scheduler, that (specifically) benefits Ryzen 3rd-Gen CPUs, with their binned cores, and thread allocation, and whatnot, during Turbo. Maybe the Qualcomm CPU/SoC is benefiting from that change as well?

I'm just uncomfortable with A/B benchmarks, with different OS versions.
 

CluelessOne

Member
Jun 19, 2015
41
21
81
32bit
MS cannot emulate x86-64 on ARMv8 for now.
So you cannot run x86-64 apps on 8cx
Thanks, for office it is not a big deal since very few really need 64 bit. But for other softwares, some really need to be 64 bit. And I am not optimistic that developers will recompile to ARM. Or they will do just the minimum.
 

soresu

Senior member
Dec 19, 2014
219
35
91
Unfortunate. Seems like the ARMy should be supporting it over NEON even for 128-bit SIMD implementations.
It's highly unlikely that the original SVE will make it into any standard (non custom) Cortex v8-A designs - the followup SVE2 was just announced a few months ago which addresses certain disparities with the NEON SIMD feature set, however the post that announced it (and TME - Transactional Memory Extensions) unequivocally stated that NEON would remain for backwards compatibility with older codebases.

Certain language used in the announcement put the SVE2 and TME developments as above the minor v8.x ISA version improvements, very possibly hinting at a major change to v9 in the future, which wouldn't be greatly surprising considering that v7-A was announced only 6 years before v8-A was in 2011, while v8-A has now been around for 8 years.

I would assume as v8-A kept v7-A software compatibility (therefore 32 bit ARM by extension), then it's likely that a v9-A profile would keep v8-A software compatibility (including NEON) in order to ensure a suitably large installed software base and take advantage of the significant work done to make ARM v8-A worthy for server deployments.

This will likely result in 32 bit ARM being dropped, and is possibly one driving reason for Google pushing devs to make the switch to ARM64 within the Play Store, Apple having already done so (though they transitioned to 64 bit v8-A cores sooner too).

Edit: It would be my prediction given the flexibility of SVE2 code that future 'little' cores will probably top out at 128 bit vector length equal to current NEON units, while future 'big' cores will likely start with 256 bit length at most, given ARM's power efficient focus.
 
Last edited:

Thala

Senior member
Nov 12, 2014
800
177
116
This will likely result in 32 bit ARM being dropped, and is possibly one driving reason for Google pushing devs to make the switch to ARM64 within the Play Store, Apple having already done so (though they transitioned to 64 bit v8-A cores sooner too).
Microsoft is currently still supporting ARM32 via WoW in Windows 10 - i wonder if this going to be dropped eventually. Personally i do not bother with 32 bit anymore when compiling for Windows.
 
  • Like
Reactions: Drazick

DrMrLordX

Lifer
Apr 27, 2000
13,015
2,688
136
@soresu

Woops I had forgotten about the distinction between SVE and SVE2. So you don't predict consumer support for SVE2 to come until ARM v9? A bit unfortunate, but I guess that's understandable.

I wonder if we'll see SVE2 deployed in ARM server cores before consumer devices?
 

Tup3x

Senior member
Dec 31, 2016
234
106
86
8cx vs Renoir? I like it. Or maybe the next generation beyond 8cx if Qualcomm is quick enough to update their platform.

Regardless, this thing is much faster than the 855. The ARMy is finally here, folks. Things could get . . . interesting.

Does anyone know if 8cx supports SVE at all?
Generation or two from now and things get really interesting. For businesses use especially.
 

soresu

Senior member
Dec 19, 2014
219
35
91
Microsoft is currently still supporting ARM32 via WoW in Windows 10 - i wonder if this going to be dropped eventually. Personally i do not bother with 32 bit anymore when compiling for Windows.
This is already where Apple is at, and where Google is pushing as of this month.

ARM64 is the future for the Cortex A-profile - though not necessarily for the more power conscious embedded M and R profiles which seem to be diverging at the OS level anyway (IoT, control systems
and such).

@soresu

Woops I had forgotten about the distinction between SVE and SVE2. So you don't predict consumer support for SVE2 to come until ARM v9? A bit unfortunate, but I guess that's understandable.

I wonder if we'll see SVE2 deployed in ARM server cores before consumer devices?
I would say SVE2 is unlikely for Cortex v8-A, and certainly TME.

The announcement blog post outlined the two as 'multi year' investments, which would seem like an odd thing to bolt onto a seemingly otherwise mature ISA, at least if there is any chance of v9-A happening.

Again there is no certainty here, they are simply pre-heating the oven as it will on these 2 big ISA changes, laying the software groundwork for the compilers prior to new software being written for them, same as Intel giving a heads up well before AVX landed to prepare the way, we know it's not inA77, and it's unlikely to be in A78/Hercules in my opinion.

What is certain from the wording of the post is that they intend SVE2 to replace NEON at the very least for 128 bit SIMD level, having shown competitive or superior predictions of performance on comparable workloads vs NEON - the only question there is of relative silicon size of the 128 bit unit..

Of course, what is less certain is exactly what those predictions/projections in the blog are based on - perhaps an otherwise useless FPGA test chip platform, or maybe a future Cortex A core as yet unnamed.
 

DrMrLordX

Lifer
Apr 27, 2000
13,015
2,688
136
What is certain from the wording of the post is that they intend SVE2 to replace NEON at the very least for 128 bit SIMD level, having shown competitive or superior predictions of performance on comparable workloads vs NEON - the only question there is of relative silicon size of the 128 bit unit..
Torvalds on ARM in the server room:

https://www.techrepublic.com/article/linus-torvalds-praises-arm-servers-but-claims-the-economics-and-ecosystem-are-missing/

He had a point, and that's that code profiling on a cheap consumer device has mostly been possible on x86 while it mostly hasn't been possible with ARM. I think 8cx and 7cx devices might change that (and let's face it, most ARM server CPUs still only support NEON). Having SVE2 in a consumer device will be even better for the ARM ecosystem since any code you write for profiling on your local machine should scale up to wider SVE2 implementations in the server room. That's something you can't do with AVX512 unless you've got Skylake-X in your box at home. Cannonlake was supposed to change that, and Icelake will for those rare few who actually get Icelake-U/Y processors, but for the majority of Intel buyers, no dice.

Of course, what is less certain is exactly what those predictions/projections in the blog are based on - perhaps an otherwise useless FPGA test chip platform, or maybe a future Cortex A core as yet unnamed.
I'm sure they've got simulations of some sort. Not really sure how they figure all that out though.
 

DrMrLordX

Lifer
Apr 27, 2000
13,015
2,688
136
Okay, that's a bit better. Good old Geekbench. I'm still surprised it isn't turning in scores closer to A12x though.
 

ASK THE COMMUNITY