Announcement

Collapse
No announcement yet.

Benchmarking The First RISC-V Cloud Server: Scaleway EM-RV1 Performance

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

  • #11
    Originally posted by c117152 View Post
    you're supposed to use specific compile flags
    It is not feasible to expect people to recompile their software to get best (or any) performance. Most will install the provided distro, such as Debian or Ubuntu and will get whatever compile flags were used by those, just like the author did. Gentoo is not nearly as popular and likely not even available on most cloud/server providers.

    Comment


    • #12
      Regarding RISC-V performance, the interesting thing on the horizon seems to be Tenstorrent's planned 128-core 8-wide-decode Ascalon processor. However it seems too early to tell if and when it will become an actual product (not just IP), and if it will also be available as a general purpose CPU (not only to drive AI accelerators).

      Comment


      • #13
        Originally posted by pWe00Iri3e7Z9lHOX2Qx View Post
        Yup. There's a reason RISC-V has already found success in micro-controllers. We are a long way away from usable laptop / desktop performance. IBM Power11+ seems like a better contender to give us something open and performant for those use cases near to mid term. I hope like hell that both succeed.
        we do not even have a opensource version of the power10 yet...

        who still buy power9 ?
        Phantom circuit Sequence Reducer Dyslexia

        Comment


        • #14
          The problem with RISC-V chips is not the ISA specifications, it's the CPU implementations that are still shy in terms of maturity/investments compared to arm/x86.

          Comment


          • #15
            I think a threadripper hypervisor managed so that 64 servers get a dedicated 2 thread core each will be much denser, and more energy efficient, and faster.

            And unless high fairness and isolation guarantees are needed, simply run the servers on virtual machines straight. Even more performance available in most situations.
            Last edited by varikonniemi; 15 May 2024, 12:29 AM.

            Comment


            • #16
              c906 was cool in 2019. We are in 2024.

              I don't think there's much to be excited about here.

              Comment


              • #17
                Originally posted by rmrm View Post
                It is not feasible to expect people to recompile their software to get best (or any) performance. Most will install the provided distro, such as Debian or Ubuntu and will get whatever compile flags were used by those, just like the author did. Gentoo is not nearly as popular and likely not even available on most cloud/server providers.
                First thing first, here's the gcc 14 testsuite for thead vector extensions mentioning using "-march=rv64gc_xtheadvector" instead of "-march=rv64gcv": https://github.com/gcc-mirror/gcc/bl...tor/pr114194.c

                With that out of the way, this is a benchmark for specific cloud hardware so I feel users will definitely make an effort to at least compile their own code with the appropriate flags. As for whether distributions decide to spin images and/or repos for rvv 0.7.1, it will depend on the longevity of the hardware and the popularity of the platform among their users.

                Regardless, like all, I hope the 0.7.1 fragmentation will die-out with in the next few years as T-Head's new core updates - which already support the ratified 1.0 vector instructions - will replace the current 0.7.1 cores. However, so long as the RISC-V platform hasn't widely adopted a universal booting mechanism like UEFI/DistroBoot and many boards require device trees and thus specific images, there's probably no harm in having additional repos providing partial coverage for popular packages for thead hardware if there willing contributors (money and maintainers) willing to take on the work. After all, the hardware WILL die out in the next 10 years or so.

                Comment


                • #18
                  RISC-V btw.

                  Comment


                  • #19
                    This is a board exposed to the public internet... It would be better compare it to other ARM boards...

                    Comment


                    • #20
                      Originally posted by rmrm View Post
                      It is not feasible to expect people to recompile their software to get best (or any) performance. Most will install the provided distro, such as Debian or Ubuntu and will get whatever compile flags were used by those, just like the author did. Gentoo is not nearly as popular and likely not even available on most cloud/server providers.
                      RISC-V platforms are still under development, and compilers are not as mature for it compared to other hardware, which means having an up to date compiler can make a difference.

                      Your comment would be more on point once RISC-V is more mainstream. Hopefully we will get there.

                      Comment

                      Working...
                      X