Ryzen keeps winning: AMD reaches 33.6% desktop share as Intel slips again

[Puiu]

Exactly, you moved them, there you go:


First, start telling where I moved goalposts? It seems you both are doing it..

Yeah? We can estimate performance M4 vs M5, agreed. But question was all the time about M4/M5 vs x86-64. Right? If you cannot compare M4 vs x86-64, there is no point comparing M5 vs x86-64 either.

All configurations? There is basically one or two configurations on M4 (and other M4 variants) and gazillion variations on x86-64.

OK, this is Cinebench: https://chipsandcheese.com/p/cinebench-2024-reviewing-the-benchmark

As for Handbrake, Zen 4 vs Zen 5 results clearly indicate AVX512 is used quite little too (as some of performance difference comes from Zen4 vs Zen5). So easily I proven you wrong.

Memory config use standard DIMMs, very different from DIMM vs LPDDR5. For some reason, there is no LPDDR5 systems on comparison.

AVX was supported even on Sandy Bridge. AVX512 is supported on Intel Alder Lake and Raptor Lake, for some reason Intel just decided to disable support. If Intel thinks those CPUs are better without it, then it's Intel decision. Not AMD problem definitely.

I know SVE. However like with AVX, AVX2, AVX512 etc question is not if it's supported but how Effectively it is supported. Zen5 and Zen5c both support AVX512 but have different effectiveness on support (5c is theoretically about half speed).

Those "benchmarks" sadly include CPUs that Cinebench team says are useless for AVX512 calculations on Cinebench. It is also around 8 years old.

And once again, software that take advantage of AVX512 are shown to have MUCH more performance gain Zen5 vs Zen4 than Blender does. That does indicate Blender does not use AVX512 effectively. Once again, it's not about Support but Effectiveness. Blender seems to be somewhat poor on that Effectiveness. And if, according to your paper, Handbrake AVX512 support is for Skylake era CPUs, it probably sucks.

Those x86 CPUs with integrated memory have very much crippled CPU compared to high performance ones. We are still talking about CPU architecture performance, right? Those CPUs are well behind what architecture is capable for.

You have yet to provide any benchmarks that disprove anything I've shown you so far. It's not that you don't want to, you can't. All you have are some limited AVX 512 workloads that don't apply to the vast majority of things you personally do on your PC and then you talk about SVE "efficiency" without anything to back up your claims.

"For some reason, there is no LPDDR5 systems on comparison." yes we do have them, both AMD and Intel use them on laptops. Did you forget about this 12 core CPU? It uses LPDDR5x-8000 RAM. And we have some youtube videos comparing Intel/AMD with Apple, the same with articles. What are you so confused about?

https://www.notebookcheck.net/AMD-Ryzen-AI-9-HX-370-Processor-Benchmarks-and-Specs.836729.0.html

It seems like all applications that consumers use and have AVX 512 support are just bad because they don't give the results you want. It's as if performance doesn't scale linearly and architectures have different bottlenecks in different workloads. it's like this sin't common knowledge for you, even though you visit techspot so often.

"Those x86 CPUs with integrated memory have very much crippled CPU compared to high performance ones" - it's because, and read it with, they can't make them to be efficient enough. And you know this to be the real reason why.

 

[HardReset]

ok, here - let me sum up just about every post you've put on this site: "None of your posts matter, AMD rules!"

happy now?

It seems trolling is great way to raise post count.

You have yet to provide any benchmarks that disprove anything I've shown you so far. It's not that you don't want to, you can't. All you have are some limited AVX 512 workloads that don't apply to the vast majority of things you personally do on your PC and then you talk about SVE "efficiency" without anything to back up your claims.

Lett me remind once again:

"Apphe has not swappable CPU and expandable memory. " - and? it's irrelevant. we are talking about CPU architectures and how fast ARM is today, not Apple's business decisions.

CPU architectures, great. Now, next you say:

You have yet to provide any benchmarks that disprove anything I've shown you so far. It's not that you don't want to, you can't. All you have are some limited AVX 512 workloads that don't apply to the vast majority of things you personally do on your PC and then you talk about SVE "efficiency" without anything to back up your claims.

Now what this has to do with CPU architectures? See, you want to measure CPU architecture speed using benchmarks, you must take benchmark that effectively takes all performance CPU architecture has to offer. For PC, you could take some 1990s software that do not support any MMX, SSE or AVX. That is then representation what modern (Zen5 for example) architecture is capable of? No. As for AVX, Cinebench team said they do not want to use AVX512 because Intel Skylake class CPUs lower clock speeds on heavy AVX512 usage scenarios and overall performance get worse. Whitepaper you showed about Handbrake AVX512 "support" clearly states performance with Skylake era CPUs. Again, heavy AVX512 usage on Skylake era CPUs mean lower performance (according to Cinebench team) so Handbrake do not heavily use AVX512, because it would mean lower performance on Skylake CPU they used.

Talking about CPU architectures, I'm not interested how "majority of software" performs or anything. For benchmarking architecture, you need at least very well optimized benchmarking scenario that you so far have not provided. You want to discuss how CPU perform on current software, fine. But that may have basically nothing to do about CPU architecture performance.

"For some reason, there is no LPDDR5 systems on comparison." yes we do have them, both AMD and Intel use them on laptops. Did you forget about this 12 core CPU? It uses LPDDR5x-8000 RAM. And we have some youtube videos comparing Intel/AMD with Apple, the same with articles. What are you so confused about?

https://www.notebookcheck.net/AMD-Ryzen-AI-9-HX-370-Processor-Benchmarks-and-Specs.836729.0.html

It seems like all applications that consumers use and have AVX 512 support are just bad because they don't give the results you want. It's as if performance doesn't scale linearly and architectures have different bottlenecks in different workloads. it's like this sin't common knowledge for you, even though you visit techspot so often.

"Those x86 CPUs with integrated memory have very much crippled CPU compared to high performance ones" - it's because, and read it with, they can't make them to be efficient enough. And you know this to be the real reason why.

I already told you why that was not included in Techspot test. I was talking about that test, but now you expand it to another tests

AVX512 performance do not scale same way on different architectures, yes. Current software do not take full advantahe of it, yes. Again, you want to talk about ARCHITECTURE performance, I'm not at all interested how today's badly optimized software works. Zen5 has proven to have excellent AVX512 performance and if current software do not use AVX512 effectively, then current software is bad for testing Zen5 ARCHITECTURE. Simple, eh?

Apple CPUs are designed: Mobile first, desktop second, HEDT third, there is no server offerings at all.

AMD CPUs are designed: Server first, HEDT second, desktop third, AMD knows people will buy Intel mobile anyway so they just don't care.

Comparing mobile first CPU against server first CPU on mobile probably favours mobile CPU? That is expected.

 

[Squid Surprise]

It seems trolling is great way to raise post count.

Lett me remind once again:

CPU architectures, great. Now, next you say:

Now what this has to do with CPU architectures? See, you want to measure CPU architecture speed using benchmarks, you must take benchmark that effectively takes all performance CPU architecture has to offer. For PC, you could take some 1990s software that do not support any MMX, SSE or AVX. That is then representation what modern (Zen5 for example) architecture is capable of? No. As for AVX, Cinebench team said they do not want to use AVX512 because Intel Skylake class CPUs lower clock speeds on heavy AVX512 usage scenarios and overall performance get worse. Whitepaper you showed about Handbrake AVX512 "support" clearly states performance with Skylake era CPUs. Again, heavy AVX512 usage on Skylake era CPUs mean lower performance (according to Cinebench team) so Handbrake do not heavily use AVX512, because it would mean lower performance on Skylake CPU they used.

Talking about CPU architectures, I'm not interested how "majority of software" performs or anything. For benchmarking architecture, you need at least very well optimized benchmarking scenario that you so far have not provided. You want to discuss how CPU perform on current software, fine. But that may have basically nothing to do about CPU architecture performance.

I already told you why that was not included in Techspot test. I was talking about that test, but now you expand it to another tests

AVX512 performance do not scale same way on different architectures, yes. Current software do not take full advantahe of it, yes. Again, you want to talk about ARCHITECTURE performance, I'm not at all interested how today's badly optimized software works. Zen5 has proven to have excellent AVX512 performance and if current software do not use AVX512 effectively, then current software is bad for testing Zen5 ARCHITECTURE. Simple, eh?

Apple CPUs are designed: Mobile first, desktop second, HEDT third, there is no server offerings at all.

AMD CPUs are designed: Server first, HEDT second, desktop third, AMD knows people will buy Intel mobile anyway so they just don't care.

Comparing mobile first CPU against server first CPU on mobile probably favours mobile CPU? That is expected.

Here's a thought... instead of posting "War and Peace" in an attempt to inundate us with nonsense... make your points clear and concise? Oh wait... I refer you back to your original (and only) point... "AMD rules!"