Announcement

Collapse
No announcement yet.

Fujitsu A64FX Support Added To The LLVM Clang 11 Compiler

Collapse
X
Collapse
 
  • Page of 1
  • Filter
  • Time
  • Show
Clear All
new posts
Previous template Next
  • #1

    Fujitsu A64FX Support Added To The LLVM Clang 11 Compiler

    Phoronix: Fujitsu A64FX Support Added To The LLVM Clang 11 Compiler

    Fujitsu has contributed support for their high-performance A64FX ARMv8-based CPU cores to the LLVM Clang compiler...

    Fujitsu A64FX Support Added To The LLVM Clang 11 Compiler - Phoronix
    http://www.phoronix.com/scan.php?page=news_item&px=Fujitsu-A64FX-LLVM-Clang-11
    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
    Tags: None
  • #2
    I hope we will see some benchmarks for the A64FX soon, as it is the first ARM processor to support the SVE extension and uses HBM2. The impact of that higher bandwith on general purpose workloads [not just vector processing] could be huge.
    • Likes 5

    Comment

    • #3
      So Michael, do you think you can get a loaner from Fijitsu? No seriously, it would be fantastic to have this machine in your labs. It will likely be awhile before Linux is fully running stable on the platform so testing would have to be revisited from time to time. So maybe 6 months to a year from now.

      A64FX is the first ARM workstation processor that has targeted performance and not web workloads. The only problem here may be the difficulty in really leverageing the vector units with common software. In any event just think about all of the fun you can have.
      • Likes 2

      Comment

      • #4
        Originally posted by wizard69 View Post
        So Michael, do you think you can get a loaner from Fijitsu? No seriously, it would be fantastic to have this machine in your labs. It will likely be awhile before Linux is fully running stable on the platform so testing would have to be revisited from time to time. So maybe 6 months to a year from now.

        A64FX is the first ARM workstation processor that has targeted performance and not web workloads. The only problem here may be the difficulty in really leverageing the vector units with common software. In any event just think about all of the fun you can have.
        Don't have any contacts at Fujitsu but I know Fujitsu uses PTS and from time to time they did upload their previous SPARC based results to OpenBenchmarking.org so at least for now that is probably the best bet where to find data.
        Michael Larabel
        https://www.michaellarabel.com/
        • Likes 3

        Comment

        • #5
          I'm sure it will appear in your favorite cloud still in this year ...

          Comment

          • #6
            According to The Next Platform, the only systems which currently use this chip are supercomputers, e.g. "Storm" line of Cray CS500. Fujitsu is looking for OEMs who want to use the chip in their own systems, so it will be awhile before we can see this chip running in a machine that an individual (or small firm) can buy.
            • Likes 2

            Comment

            • #7
              Originally posted by pegasus View Post
              I'm sure it will appear in your favorite cloud still in this year ...
              Maybe not your favorite cloud, but perhaps some public cloud platform. I don't see Amazon using these and probably not Google.

              Originally posted by Bronek View Post
              According to The Next Platform, the only systems which currently use this chip are supercomputers
              It's a good point. They have a lot of features & HW that probably isn't applicable to most cloud workloads.

              Comment

              • #8
                Originally posted by coder View Post
                Maybe not your favorite cloud, but perhaps some public cloud platform. I don't see Amazon using these and probably not Google.


                It's a good point. They have a lot of features & HW that probably isn't applicable to most cloud workloads.
                Yes, these chips have a curious balance of 32GB HBM2 (and no support for attachable RAM!) and 48 cores (+ 4 spare!). I am sure that there are workloads where this balance would work very well, but with less than 1GB per core this is definitely not your typical cloud workload. We are talking intensive compute tasks with high degree of parallelism, which is not how general purpose software is typically written. However, IMO a big cloud provider or a hyperscaler might be still interested because hey, for some workloads this just might be the right kind of machine. I guess we should see next year.
                • Likes 1

                Comment

                • #9
                  Originally posted by coder View Post
                  Maybe not your favorite cloud, but perhaps some public cloud platform. I don't see Amazon using these and probably not Google.
                  Yes, but Amazon have this :
                  Cost Analysis - An x86 Massacre
                  https://www.anandtech.com/show/15578...ntel-and-amd/9
                  Last edited by Skwaloo; 10 March 2020, 06:15 PM.
                  • Likes 1

                  Comment

                  • #10
                    Originally posted by Skwaloo View Post
                    Yes, but Amazon have this :
                    Cost Analysis - An x86 Massacre
                    https://www.anandtech.com/show/15578...ntel-and-amd/9
                    I've had an opportunity to talk with guys at Arm recently and they are very smart, but still the scale of this .... that comes as a surprise.

                    Comment

                    Previous template Next
                    Working...
                    X