Intel Battlemage Showing Off Nice OpenCL Gains With Newest Open-Source Compute Stack

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts
  • phoronix
    Administrator
    • Jan 2007
    • 67406

    Intel Battlemage Showing Off Nice OpenCL Gains With Newest Open-Source Compute Stack

    Phoronix: Intel Battlemage Showing Off Nice OpenCL Gains With Newest Open-Source Compute Stack

    Last month with the launch of Intel Battlemage with the Arc B580 graphics card, there was fairly nice open-source GPU compute performance but with some outliers... Today it's a pleasure to report that with the newest open-source GPU compute stack as of this past week, there are some nixe Xe2 / Battlemage improvements for enhancing the performance of some OpenCL workloads and also correcting the performance of some workloads that were in poor standing on launch day.

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite
  • schmidtbag
    Senior Member
    • Dec 2010
    • 6618

    #2
    Those are some great improvements. Some of those SHOC results are insane - almost makes me wonder if it's even working correctly.

    Comment

    • Jumbotron
      Senior Member
      • Jul 2015
      • 1271

      #3
      Originally posted by schmidtbag View Post
      Those are some great improvements. Some of those SHOC results are insane - almost makes me wonder if it's even working correctly.
      I’ll say !! The clpeak kernel latency drop is beyond insane !

      Comment

      • Quackdoc
        Senior Member
        • Oct 2020
        • 5116

        #4
        aways happy to see performance increases. Hope we see more regarding vulkan compute specifically however. Vulkan compute is quickly beccoming more popular as things like wgpu abstract over vulkan making it more and more competitive and a "good option".

        For instance, there is a PR someone made to add wgpu to candle, huggingface's minimal ML framework. Or traceai's burn stuff. Candle work is pretty great, I got whipser working flawlessly on my a380 (though the lack of f16 in wgpu hurts, we are seeing that change hopefully this week) Aside from AI stuff, we already see vulkan being used for upscalers in video processing as well.

        Comment

        • JEBjames
          Senior Member
          • Jan 2018
          • 377

          #5
          Michael

          Typo page 1

          "some nixe" should be "nice"

          Comment

          • Anon'ym'
            Phoronix Member
            • Jul 2021
            • 59

            #6
            Strange that BM GPU making really good in double precision test.
            Maybe it is beneficial to use 32bits primitives for this cards. When code for consumer cards usually optimized for 16bit?

            Comment

            • coder
              Senior Member
              • Nov 2014
              • 8968

              #7
              Nice to see how well it's comparing against the A770, but Intel didn't make such good progress against one of my biggest issues with that card: power consumption. The B580 idles way too high and uses way too much power in simple tasks like video playback, at least during this Windows-based testing:

              The next-generation Intel Arc graphics cards are here! The B580, powered by the Battlemage architecture, is priced at a highly competitive $250. Testing in our review confirms that Intel's new card outperforms both NVIDIA's GeForce RTX 4060 and AMD's RX 7600, and it now supports frame generation as well.


              Also, good job not only bringing back fp64 compute, but also punching above its weight on that front!

              Comment

              • coder
                Senior Member
                • Nov 2014
                • 8968

                #8
                Originally posted by Quackdoc View Post
                we already see vulkan being used for upscalers in video processing as well.
                Such as?

                Comment

                • coder
                  Senior Member
                  • Nov 2014
                  • 8968

                  #9
                  Originally posted by Anon'ym' View Post
                  Maybe it is beneficial to use 32bits primitives for this cards. When code for consumer cards usually optimized for 16bit?
                  WTF?

                  In general, desktop GPUs didn't really embrace fp16 until AI became popular. I'm sure game engines are still using fp32 compute for most things. I think fp16 only caught on for mobile GPUs, like in phones.

                  Not only that, but I also don't see what any of that has to do with their fp64 performance!

                  Comment

                  • Anux
                    Senior Member
                    • Nov 2021
                    • 1960

                    #10
                    Originally posted by coder View Post
                    In general, desktop GPUs didn't really embrace fp16 until AI became popular.
                    I'm fairly certain it's already in use for shader programming in modern games.

                    Comment

                    Working...
                    X