RISC-V Made Nice Software Progress In 2024 While Interesting Hardware Still Rare

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

    RISC-V Made Nice Software Progress In 2024 While Interesting Hardware Still Rare

    Phoronix: RISC-V Made Nice Software Progress In 2024 While Interesting Hardware Still Rare

    RISC-V on the software front made very nice progress over the past year with a lot of Linux kernel and toolchain improvements, new targets being enabled, and new instructions being supported along with other additions for improving the overall RISC-V software ecosystem. When it comes to hardware though most of the readily available RISC-V systems are painstakingly slow and the more performant/featureful options are much harder to come by...

    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
  • ayumu
    Senior Member
    • Oct 2008
    • 673

    #2
    Debian sid build status: https://buildd.debian.org/status/arc...cv64&suite=sid

    Relative to other arch: https://buildd.debian.org/stats/graph-week-big.png

    Not bad at all. From nothing, to the current point, about to pass ppc64 to become the third architecture in built debian packages.

    2025 is just starting. It's not crazy to think it could pass arm64 by the end of the year.

    Comment

    • Quackdoc
      Senior Member
      • Oct 2020
      • 5116

      #3
      I think the milkv mars will be a mainstay for a while, it's fairly low power, but has really good performance for the wattage of it, makes it a real competitor to the rpi3b. it's a bit of a shame the new sipeed machine didn't launch before the new year, but IMO it will be the most interesting SBC of 2025. mainly because the SOM modules already have some decent hardware to go into, it would be nice to see other risc-v manufactures make soms compatible with them.

      Comment

      • rmfx
        Senior Member
        • Jan 2019
        • 765

        #4
        I think SiFive is a bit disappointing. They have all the cards in their hands, yet, they struggle to get good stuff out.
        Their latest dev board is nice but still not based on their latest core tech.

        And imho, they should release alongside a Raspberry Pi competitor themselves at a contained price.

        Comment

        • Guest

          #5
          Originally posted by ayumu View Post

          2025 is just starting. It's not crazy to think it could pass arm64 by the end of the year.
          I will say that the whole problem is not what you all think. I do not like RISC-V, because of the lie embedded in the foundation of this architecture. But even now this RISC-V could surpass everything that exists. Let's take RISC-V cores and make their arrangement and interaction with each other in such a way that their weak points become strengths. Their instruction set and so on should be used in such a way that it would be possible to do something that will give real parallel execution of tasks. Such a meta architecture can be built on any cores. But if you like RISC-V then it is possible to do all this in a short time. You just need a clear understanding of things, which you do not have here. You just want money. Make a mate architecture based on a large number of RISC-V cores and adapt at least one programming language for this, and you will succeed


          Comment

          • Quackdoc
            Senior Member
            • Oct 2020
            • 5116

            #6
            Originally posted by rmfx View Post
            I think SiFive is a bit disappointing. They have all the cards in their hands, yet, they struggle to get good stuff out.
            Their latest dev board is nice but still not based on their latest core tech.

            And imho, they should release alongside a Raspberry Pi competitor themselves at a contained price.
            watchu talkin about willis? The p550 that they are releasing *is* the new stuff, the P650 hasn't been manufactured yet afaik. You still need to design the SOC around the core. these things take a LOT of time, the only reason why it seems so long is because there is a good amount of press/openness around it. Also it's not sifive's job to compete with RPI, leave that to people like milkv or sipeed that are better equipped to do so, and are doing so.

            The milkv mars competes directly with the rpi3b, and the lichee pi4a competes with something like the rockchip devices, and the upcomming liche pi5a andmilkv megrez will be likely in a good position to compete with the rpi5, we can see some initial performance and usability here https://www.youtube.com/watch?v=9KTbi8dJjzQ

            Comment

            • rmfx
              Senior Member
              • Jan 2019
              • 765

              #7
              Originally posted by Quackdoc View Post

              watchu talkin about willis? The p550 that they are releasing *is* the new stuff, the P650 hasn't been manufactured yet afaik. You still need to design the SOC around the core. these things take a LOT of time, the only reason why it seems so long is because there is a good amount of press/openness around it. Also it's not sifive's job to compete with RPI, leave that to people like milkv or sipeed that are better equipped to do so, and are doing so.

              The milkv mars competes directly with the rpi3b, and the lichee pi4a competes with something like the rockchip devices, and the upcomming liche pi5a andmilkv megrez will be likely in a good position to compete with the rpi5, we can see some initial performance and usability here https://www.youtube.com/watch?v=9KTbi8dJjzQ
              P650 is not their latest core either, they have their P870 and even newer the P870D.

              The promise from SiFive was a much better timeline from design to silicon, it seems they are not fulfilling this goal yet clearly. It may take time to build a SOC design but these things can be done in parallel too, it’s not serial.

              Comment

              • Quackdoc
                Senior Member
                • Oct 2020
                • 5116

                #8
                Originally posted by rmfx View Post
                P650 is not their latest core either, they have their P870 and even newer the P870D.

                The promise from SiFive was a much better timeline from design to silicon, it seems they are not fulfilling this goal yet clearly. It may take time to build a SOC design but these things can be done in parallel too, it’s not serial.
                Don't forget that SBCs get old tech, we rarely get anything as soon as is "ready". normally we get special use cases first (IE. antminer vs pioneer) and then it finally trickles down to consumer products. But even with this they are already operating insanely fast. the P550 was announced in june 2021, horsecreek was supposed to launch early 2022, but was delayed by intel, and then finally intel dropped out in 2022/2023 sometime. it dropped off the face of the news until 2024 when it was announced the SOC was going to be made by eswin. IMO it's fairly safe to assume that there was a solid time frame there when development was just out right not happening. But even then 4 years is still a good time frame to go from announcement to a purchasable product, especially since you still need to design the actual SBC around the SOC.

                The same thing is now potentially happening with the P670, Sophon/sophgo, can't remeber which specfically, was supposed to be working on the SG2380, but that may have went tits up dude to their parts being found in huawei devices meaning they may be barred from using TSMC which afaik was going to be fairly critical for this. It's possible that the plans have been put on halt.

                as for the P870/P870D, IIRC the tape out was only expected for end of december, I wouldn't expect anything even remotely consumer purchasable for the next 3 years assuming they hit their tapeout deadline.

                Comment

                • ayumu
                  Senior Member
                  • Oct 2008
                  • 673

                  #9
                  Originally posted by Quackdoc View Post
                  I think the milkv mars will be a mainstay for a while, it's fairly low power, but has really good performance for the wattage of it, makes it a real competitor to the rpi3b. it's a bit of a shame the new sipeed machine didn't launch before the new year, but IMO it will be the most interesting SBC of 2025. mainly because the SOM modules already have some decent hardware to go into, it would be nice to see other risc-v manufactures make soms compatible with them.
                  Milk-V Mars is a discount VisionFive 2, which was the first with that SoC and I understand the board with the most units sold. That SoC will stay relevant for some time, because it was the first RVA20-compliant mass-market RISC-V SBC.

                  This 2025 we should see a bunch of RVA22+V machines, much better than the spacemiT K1/M1 ones already available.

                  Comment

                  • ayumu
                    Senior Member
                    • Oct 2008
                    • 673

                    #10
                    Originally posted by rmfx View Post
                    The promise from SiFive was a much better timeline from design to silicon, it seems they are not fulfilling this goal yet clearly.
                    I am sure they won't make the mistake of trying to work with Intel again, and this timeline optimization will actually materialize.

                    From what I've heard, a P670 devboard might show up later in this very 2025.

                    Comment

                    Working...
                    X