No announcement yet.

SiFive HiFive Unmatched Hands-On, Initial RISC-V Performance Benchmarks

  • Filter
  • Time
  • Show
Clear All
new posts

  • SiFive HiFive Unmatched Hands-On, Initial RISC-V Performance Benchmarks

    Phoronix: SiFive HiFive Unmatched Hands-On, Initial RISC-V Performance Benchmarks

    A few weeks ago I finally received the HiFive Unmatched from SiFive as their flagship RISC-V development board. As a reminder this is their mini-ITX development board that is powered by their U740 SoC and features 16GB of DDR4 system memory, one PCI Express x16 slot that can work with AMD Radeon graphics cards on Linux, and other features. It's been a delight playing with this developer platform and enclosed are some early benchmarks as well showing off the U740 performance as well as how the Linux software support/performance has been evolving.

    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

  • #2
    Wow, that the $70 pi400 significantly outperforms this nearly 10x more expensive Risc V board says quite a lot. I'm all for interesting non-x86 uarch, but it appears Risc-V has a looong way to go to become a viable option. With 16 GB, x16 PCIe, and an NVMe M.2 slot, this board does have some compelling expansion options as compared with rpi. Do they offer a case for it, for desktop use?
    Last edited by torsionbar28; 24 September 2021, 10:06 AM.


    • #3
      It’s a mITX board so any SFF case and power supply will do. The fan is very noisy, however, and at least with nVidia GPUs the UI is slow as molasses.

      RISC-V has the full backing of the Chinese, who are looking to remove their dependency on US technology (or companies subject to US export bans like ARM). You can expect it to progress quickly but in the meantime anyone who wants a developer workstation for the platform only really has this board as a contender. I suspect QEMU emulating RISC-V on a modern x64 will be faster, though.


      • #4
        $700 for this? What a fucking joke!


        • #5
          Get the price down to the 200-300 range, better software support and this would be a cool dev board!


          • #6
            Complaining about the price of this board is a bit misguided IMHO. This piece of equipment is expensive because it is more or less a prototype manufactured in tiny quantities compared to x86 or rpi. It should not be judged based on its performance per dollar ratio. Another aspect making this more expensive is the RAM, NVMe, PCIe etc. This is not meant for hobbyists. This is meant for people who are in charge of getting other hardware or platforms up and running on RISC-V otherwise putting a dedicated GPU and an NVMe drive on such a low performance device makes no sense. To them, the price and performance are just a side note.


            • #7
              Originally posted by Mario Junior View Post
              $700 for this? What a fucking joke!
              For a dev board it's quite cheap. For example an ARM Juno costs $10000, ARM MPS3 costs $4000, ARM Neoverse N1 SDP costs $10000 +tax/shipping.


              • #8
                To be honest, I'm quite shocked at how powerful it is. You have to remember, it's basically the first "really usable" RISC-V CPU made by a completely new player in the industry. Bravo!

                Though Linux is probably quite heavy for it, so a smaller and more optimized OS like HaikuOS, SerenityOS (once it becomes usable), GenodeOS (if you can somehow navigate it) or others is probably a better fit than trying to run GTK/QT.

                Or maybe Alpine + a light GUI like sway?


                • #9
                  The Ubuntu version performance difference is probably clock rate. The first Ubuntu release accidentally grabbed the wrong bootloader version and is running at 1.0GHz. The second Ubuntu release probably has the right boot loader version and is running at 1.2GHz. If you have perf, then "perf stat /bin/ls" will give a reasonable estimate of the clock rate.


                  • #10
                    Originally posted by numacross View Post
                    Holy macaroni!