Announcement

Collapse
No announcement yet.

Intel Xeon Platinum 8490H "Sapphire Rapids" Performance Benchmarks

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Mark Rose
    replied
    Originally posted by aaahaaap View Post
    Wow, that's a pretty bad showing for Intel compared to AMD, not sure why the article is so positive.
    Because the new CPUs are better for some applications. They're a massive improvement over older generations of Xeons. They're not universally better. The review is fair.

    If Michael were unfair Intel wouldn't send him a $40,000 machine.

    Leave a comment:


  • crudeboil
    replied
    aaahaaap

    ​It's a huge improvement from Ice-Lake. But yes, this article is focused too much on AI inference. No DB benchmarks for example.

    On one hand I can understand that because AMX is new, on the other hand I can see this as a result of Intel marketing their "accelerators" for months.

    It's not like SPR is going to make a real difference for AI inference because there are two real world scenarios for inference: 1. You really need to be fast and get a GPU or dedicated AI accelerator. 2. You just do inference here and there for small models where the performance of most server CPUs will be enough. The area between these two is probably very small.

    Michael writes "at least in AI workloads with AMX and other workloads able to take advantage of new Sapphire Rapids features, the price can actually be justified": Having two $17k CPUs in your system just to do the job of a $1600 GPU slower and with much higher power consumption doesn't make much sense. I would like to see some benchmarks comparing SPR to an old GPU like the T4.

    Most important: I'm pretty sure that Intel will drop AMX soon and integrate inference accelerators into Xeon chips as dedicated circuitry, not into cores. This approach is very inefficient from power and die area perspective. Intel will have dedicated AI accelerator circuitry in their next mobile CPU, AMD has announced those for Epyc and their APUs and Apple has their Neural engine while Qualcomm has their NPUs. Everyone can see how efficient those are.

    Leave a comment:


  • aaahaaap
    replied
    Wow, that's a pretty bad showing for Intel compared to AMD, not sure why the article is so positive.

    Leave a comment:


  • onlyLinuxLuvUBack
    replied
    Originally posted by Mark Rose View Post
    It's nice to see Intel get some competitiveness back!
    they didn't call it xeon RAPID but instead xeon MAx because they knew it isn't really rapid vs amd

    Leave a comment:


  • coder
    replied
    Too bad this didn't launch 1 year ago. It holds up pretty well against the Zen3-based (Milan) EPYC. Even the X (3D cache) version!

    Leave a comment:


  • nicalandia
    replied
    Its Very Clear that OpenVIMO is making the Overall Geomean look more Competitive for Sapphire Rapids. OpenVIMO is using AVX2/AVX512, AMX. That's great for specific workload, but how easy or feasable would be to make an appp that takes Advantage of SSE/AVX/AVX2/AVX512 instructions take advantage of AMX? For example Blender.

    Here is a comparison I put together of AMD Genoa AVX-512(Anabled and Disabled) vs Intel 8490H AMX
    1673394404106.png

    Geomean
    1673394499793.png

    AMD Genoa AVX-512 ON/OFF vs Intel SPR-SP AMX​
    Last edited by nicalandia; 11 January 2023, 09:52 AM.

    Leave a comment:


  • rlkrlk
    replied
    Rather curious why some of the OpenVINO benchmarks are showing superlinear scaling on the 8490 (they are close to perfectly linear on the other chips). Any thoughts?

    Leave a comment:


  • Michael
    replied
    Originally posted by nicalandia View Post

    The 2S Genoa 9374F(64C total) is matching 2S 8490H(120C/240T)
    image.png​

    https://openbenchmarking.org/vs/Proc...atinum%208490H
    I wouldn't pay too much attention to that page, yet, due to its limited scope. It's only showing data where there is sufficient collection for each benchmark/configuration and other factors that refines it to only showing sound generalized results. As more benchmarks of the 8490H and other SPR results appear in the days/weeks/months ahead it will be much more relevant.

    Leave a comment:


  • Paradigm Shifter
    replied
    So Intel does well in benchmarks only Intel can do well in because it uses a new instruction set which only Intel currently have.

    Colour me shocked.

    But that power consumption.

    And the lack of competitiveness for non-AMX workloads (especially considering price!)

    Leave a comment:


  • nicalandia
    replied
    Originally posted by StephenL View Post
    Good article. But I think Michael you were being very charitable with the comments. A single 9654 was competitive with the 2P 8490 in most non-Intel benchmarks at a fraction of the price.

    Yes, very interested to see how the accelerators perform, and for specific AMX workloads they may be good, but for the price and power consumption, Intel still has a long way to go.
    The 2S Genoa 9374F(64C total) is matching 2S 8490H(120C/240T)
    image.png​

    https://openbenchmarking.org/vs/Proc...atinum%208490H
    Last edited by nicalandia; 10 January 2023, 07:22 PM.

    Leave a comment:

Working...
X