Announcement

Collapse
No announcement yet.

SiFive HiFive Unmatched RISC-V Developer Boards Begin Shipping

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

  • #21
    Originally posted by brucehoult View Post

    Where did you find Pentium III figures? I looked but failed so I'm going off general principles.

    I believe Pentium III (like PowerPC G4) was actually several different micro-architectures with the later 900 MHz and 1 GHz ones being better. Also testing with a current compiler would probably produce better results than a compiler from 2000.

    I've just pulled a 1.25 GHz G4 Mac Mini out of storage to try some comparisons. I don't have an Intel machine of that vintage (and I was using Athlon from 700 MHz to "3200+" anyway)


    For DMIPS, includes a fair number of P6 (Pentium 3/ Pentium M) representatives. They also have a Pi 3 on there.
    Funny to see how poor netburst/P4 actually was per MHz, and you see the huge improvement for core 2.

    https://www.eembc.org/coremark/scores.php (search for Pentium)

    Yes a more modern compiler might to a little better (especially if it vectorizes some of the loops w/ SSE).

    https://bebop.cs.berkeley.edu/resour...ltradeoffs.pdf

    A good overview of the P6 architecture. If made today with a comparable front-end, it would be fairly close to an A72/Pi 4. But keep in mind a smaller memory window, longer pipeline, worse branch predictor, and small TLB (16 entry instruction+ 32 entry data, with no L2 TLB supporting it) - Edit - Also keep in mind a better bypass network. All these small details add up to significant improvements.

    Edit - The claimed scores for the U74 RISC-V core https://www.sifive.com/cores/u74
    Last edited by WorBlux; 26 May 2021, 07:29 PM.

    Comment


    • #22
      Nice. Everyone knew at the time the P4 was just a marketing exercise "look we've got more MHz" not an actual performance leader.

      I'm a little surprised the A53 in the Pi3 is so equal to P III Tualatin with 32 bit code at the same MHz. The P III is wider, but as you say other factors have improved since then. And Dhrystone is a rubbish benchmark, but what can you do? I reckon my own Primes benchmark is a better indicator of core and L1 performance :-) I've also run that on quite a few different things: http://hoult.org/primes.txt

      The Arm A55 and SiFive U74 are markedly better dual-issue in-order cores than A53 as both have "early" and "late" ALUs that enable two dependent operations to dispatch together if the dependencies of the first instruction are already available. This increases the dual-issue percentage by a significant amount, without expensive OoO machinery.

      Comment


      • #23
        Originally posted by phoronix View Post
        this very interesting RISC-V developer board is now shipping to customers...
        Got mine. Works ok out of the box, but it doesn't see the Intel M.2 WiFi card I put in, the preloaded kernel doesn't have drivers for most USB WiFi dongles I've tried, and the AMD Radeon RX 550 card I dropped into the PCI slot doesn't work with the distro that comes with the board.
        In better news, all the onboard bits and the XPG Gammix S5 2TB M.2 SSD I've put in work perfectly.

        I've restarted the OpenMandriva RISC-V port on the board -- should have something nice that will support all the HW I've added soon-ish...

        Comment


        • #24
          Originally posted by berolinux View Post
          the AMD Radeon RX 550 card I dropped into the PCI slot doesn't work with the distro that comes with the board.
          Isn't accelerated, or doesn't work at all?

          I'd hope any video card would at least work in straight VGA mode.

          Mine says "Out for delivery", but that usually just means it's heading from the nearest city to the village where the rural delivery guy will get it from tomorrow. I have an R5 230 waiting but also just ordered an R9 350 that is in stock and shipping today.

          Comment


          • #25
            Originally posted by brucehoult View Post
            Isn't accelerated, or doesn't work at all?
            Doesn't work at all, black screen and Xorg process eating up 100% CPU.
            But the kernel module loads and dmesg output looks good.
            Probably not all that much missing.
            Unfortunately rather busy with work right now, so won't get to debug this for a bit. But fortunately work does include some related bits and pieces like getting risc-v toolchains up.
            Fortunately around here, stuff from the city arrives in the village the same day. Small countries have some advantages.

            Comment


            • #26
              Originally posted by berolinux View Post
              Fortunately around here, stuff from the city arrives in the village the same day. Small countries have some advantages.
              And it turned up 45 minutes ago :-)

              Comment

              Working...
              X