SiFive HiFive Premier P550 RISC-V Price Lowered, Ubuntu 24.04 Support Ready

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts
  • Quackdoc
    Senior Member
    • Oct 2020
    • 5072

    #21
    Originally posted by brucehoult View Post
    That's because most Geekbench tests use Vector/SIMD processing in some way. The P550 doesn't have any form of SIMD, so it's kind of silly to even run (scalar versions of) such tests. OF COURSE it will be slow. That's not useful information.
    I don't think it's necessairly unfair, nor is it not useful information in general to compare. It just happens that geekbench itself is a bad tool to use to compare, since it's non-indicative of real world perf.​ That being said, I don't really know how good various compilers are at auto vectorization yet. It may be a non-issue at all for apps without hand written asm, dunno.

    Originally posted by SteamPunker View Post
    If the P550 takes off too much, then common RISC-V distros and software packages might end up being compiled without support for any vector extensions, even though later RISC-V systems will all have support for them. And this time, we won't have a 32-bit/64-bit transition opportunity, which would allow us to start from a clean slate with determining a "base profile" for compatibility.
    this is already the status quo. If you want a distro with full rvv, gentoo it xD

    Comment

    • brucehoult
      Phoronix Member
      • Aug 2019
      • 99

      #22
      Originally posted by SteamPunker View Post

      That's also my concern about this P550 platform.

      Whether you're adventurous enough to start using a RISC-V platform as a daily driver, or would like to use it for furthering development of the RISC-V software ecosystem, at this point surely you'd want to go for a CPU that supports the recently adopted RVA23 Profile?
      Dude, that spec was JUST PUBLISHED a month or so ago. It will be 2-3 years before there is hardware for sale implementing it.

      Maybe this is part of the reason why SiFive is is discounting these boards, to clear inventory for a next-gen product that does support the RVA23 profile.
      That is ridiculous. The is THE NEWEST RISC-V SoC and board on the market. It is only three years since the P550 core was announced.

      That is a year faster than other recent Arm and RISC-V cores have taken to get into SBCs e.g. Arm A53 (Pi 3), A72 (Pi 4), A76 (Pi 5), SiFive U74 (VisionFive 2 and others), THead C910 (Lichee Pi 4A, Milk-V Pioneer) all took 4+ years from release of the core's RTL to boards for sale.

      If the P550 takes off too much, then common RISC-V distros and software packages might end up being compiled without support for any vector extensions, even though later RISC-V systems will all have support for them.
      Like every other RISC-V SBC on the market, except for the ones with SpacemiT SoC that started to ship around August this year: Banana Pi BPI-F3, Milk-V Jupiter, Sipeed Lichee Pi 3A, DC-Roma II. That is the *only* RISC-V SoC with multi-core, multi GB RAM etc that has RVV 1.0.

      People are going to be using RV64GC boards for the next 5+ years, even if these brand new P550 boards didn't exist. Software will simply have to cope with some machines having vector and some not.

      Comment

      • brucehoult
        Phoronix Member
        • Aug 2019
        • 99

        #23
        Originally posted by Quackdoc View Post
        I don't really know how good various compilers are at auto vectorization yet.
        No compiler in the world, for any ISA, can auto-vectorise worth a damn. Only the simplest loops will work, and even then they won't be the speed of hand-written code because the compiler had to be conservative and assume all kinds of things might happen than the programmer knows very well won't happen.

        Comment

        • Quackdoc
          Senior Member
          • Oct 2020
          • 5072

          #24
          Originally posted by brucehoult View Post
          No compiler in the world, for any ISA, can auto-vectorise worth a damn. Only the simplest loops will work, and even then they won't be the speed of hand-written code because the compiler had to be conservative and assume all kinds of things might happen than the programmer knows very well won't happen.
          Auto vectorization isn't superb or anything, but it does get the job done in a lot of cases, its certainly a tool that can help.

          Comment

          • microcode
            Senior Member
            • Mar 2013
            • 2374

            #25
            Is it just me, or does arrow.com just not load? They give me Access Denied no matter what, even in a stock Chrome browser. Maybe they block Linux UAs?

            Edit: yeah, looks like arrow.com literally just blocks any Linux chrome User Agent... and now it is also blocking me with the Windows UA? wtf lol
            Last edited by microcode; 12 December 2024, 12:55 PM.

            Comment

            • Gamer1227
              Phoronix Member
              • Mar 2024
              • 68

              #26
              Originally posted by SteamPunker View Post
              If the P550 takes off too much, then common RISC-V distros and software packages might end up being compiled without support for any vector extensions, even though later RISC-V systems will all have support for them. And this time, we won't have a 32-bit/64-bit transition opportunity, which would allow us to start from a clean slate with determining a "base profile" for compatibility.
              maybe, Vector extensions have been ratified in 2021, and hardware only shipped. in 2024 with the final version.

              Just like most software only started using SSE2 in ~2010, since AMD only got SSE2 many years after Intel. (2000 vs 2004)

              But today using SIMD libraries instead of creating your own library is way more common, and they normally use run time detection.

              Comment

              • Quackdoc
                Senior Member
                • Oct 2020
                • 5072

                #27
                Originally posted by microcode View Post
                Is it just me, or does arrow.com just not load? They give me Access Denied no matter what, even in a stock Chrome browser. Maybe they block Linux UAs?

                Edit: yeah, looks like arrow.com literally just blocks any Linux chrome User Agent... and now it is also blocking me with the Windows UA? wtf lol
                wtf, this is absolutely wild, someone probably messed up a config... I hope

                Comment

                • coder
                  Senior Member
                  • Nov 2014
                  • 8952

                  #28
                  Originally posted by brucehoult View Post
                  No compiler in the world, for any ISA, can auto-vectorise worth a damn.
                  It depends on your level of expectation. It won't refactor your control and datastructures to make your code vectorization-friendly. However, if you put in that prep work (which you'd have to do anyway, if you want to hand vectorize it worth a damn) and you know how to use the restrict keyword and when it makes sense to use -Ofast, you can get decent results. If you give it some hints, either using builtins, or with PGO, then your results will be even better.

                  The main problem people run into is that they see the compiler flounder at autovectorization and arrogantly assume that it's just stupid, rather than asking why it didn't do what they expected and trying to figure out what it got hung up on. If you put in the effort to do that, then you might be surprised at how good they can be and you might save yourself some time & effort from having to do the entire process by hand.

                  Originally posted by brucehoult View Post
                  ​even then they won't be the speed of hand-written code because the compiler had to be conservative and assume all kinds of things might happen than the programmer knows very well won't happen.
                  It's the programmer's fault for not telling the compiler those things won't happen. The language is well-defined and the tools give you nearly all the expressiveness you need to say what you actually mean. If you fail to tell the compiler what you really mean, that's on you, not the compiler.

                  Seriously, do you hear machinists blaming their milling machines for being dumb and advising to do the work by hand? No! They learn how to use their tools and we should do the same!

                  Comment

                  • varikonniemi
                    Senior Member
                    • Jan 2012
                    • 1102

                    #29
                    Originally posted by NateHubbard View Post

                    There's the guy that's so much fun at parties.
                    We are indeed having fun laughing at exorbitant launch prices that had to be severely cut when the profit gouging leadership had a meeting with reality. Come join the ones laughing so you don't have to keep defending the crying price gougers to the grave.

                    Comment

                    • bog_dan_ro
                      Junior Member
                      • Apr 2015
                      • 12

                      #30
                      The price drop has nothing to do with milk-v megrez, right ?

                      Comment

                      Working...
                      X