top500-benchmark

Nice and organised work! My only reservation wrt the core of it is that it tests the cpus on a matrix operations task, and power efficiency in this case refers specifically to doing matrix operations. This is fine if that's the kind of task one wants to optimise for, but it does not necessarily translate to general energy efficiency for a broader spectrum of tasks that most people may do most of the time. And thus, if one wants to optimise for matrix operations, wouldn't it make sense to use apple's accelerate framework instead of blis/openblas (I think you used blis?)? I compiled hpl to use accelerate [0] on my m1 max (macbook), and ran it with settings almost identical to the original m1 max settings [1]. I get 422 Gflops vs 264 Gflops in m1 max mac studio (docker) [2]. My average draw (using a wall power meter, but not interfaced with

Not only that, the efficiency (FP64, HPL) is also beating all the M4/M3/M2/M1 systems I've tested besides the M4 Mac mini: https://github.com/geerlingguy/top500-benchmark#results I think the bigger news is Intel's made a chip that isn't just pushing clocks to hit some performance number, they've actually tuned clocks for efficiency... which also leads to much better battery life. And the fan on the system in the original article[1] also allows the Intel system to run substantially faster for longer tasks (where the Neo thermally saturates and has to stay throttled). (The iGPU is another matter... sadly still a bit lacking.) I'm hoping Intel makes a follow-up to the N100-series chips with just four of their newer low power E-cores. [1] https://www.jeffgeerling.com/blog/2026/excited-for-intel-eff...

All his tests are published here: https://github.com/geerlingguy/top500-benchmark?tab=readme-o... He does test some N100 machines. TL;DR: The M4 blows everything away, but isn't a general purpose machine. The N100 is slightly better than 50% of the power efficiency of the remainder. It's also slightly faster despite having less cores. Single core speed is looks to be twice the speed of ARM. An N100 box with 16GB of RAM and 1TB of NVME is around the USD$200 mark, which is far cheaper than the Mac or Ampere, but in line with the other ARM options. It comes in more form factors with more customisability options than you can poke a pointed stick at. All in all, it doesn't fair too badly. The low price, fast single core speed, and compatibility with everything makes up for a lot of sins.

EPYC is decent, Xeon has a long way to go still. You mention actually measuring it, that's exactly what I've been doing in my top500 project: https://github.com/geerlingguy/top500-benchmark#results The RK3588 gets up to 4.64 Gflops/W, only beaten by a couple expensive Ampere Altra chips in my testing so far. A few of the newest EPYC chips touch that range, but none of the Xeons I've seen do (yet). I would like to test a few more but getting access to bare metal is tough, especially since I only need it for a day or two, but need to get it plugged into a consistent power monitoring device.

Note that this board runs RK1's (which use the RK3588 SoC), not Raspberry Pis. It burns 18.5W with CPU cores maxed out, producing 224 Gflops of compute (see: https://github.com/geerlingguy/top500-benchmark/issues/27). Efficiency-wise, it's a little more power efficient than at least last-gen AMD/Intel chips, and a little less than Ampere CPUs and Apple's M1. I haven't had a chance to compare to M2/M3 for power efficiency yet. The Turing Pi 2 has a BMC that allows out-of-band management, and the firmware is fairly stable at this point, and the community helped build a nice CLI for it so IMO it's even nicer (from automation/remote sysadmin aspect) than IPMI for some things. There's no remote VNC/display, but these aren't meant to have displays (even though node 1 has HDMI routed to a connector on the back). The most crippling thing for my u

HPL Linpack was tested on a Q64-22 CPU ~650 Gflops https://github.com/geerlingguy/top500-benchmark/issues/19 Memory bandwidth, it's 8 channels of DDR4 on the Ampere vs 7960x 4 channels of DDR5. AMD likely has GFLOPS in the low 1000s if I had to guess, I cannot find any benchmarks of the current gen CPUs, only last gen. I think the Threadripper is actually more expensive, it seems to be a bit less power efficient. Probably similar memory bandwidth due to higher ddr5 clock, faster at math. The PRO threadrippers are even more. It seems competitive, at least, price and performance.

Caveat: I consider myself a noob at benchmarking, after spending probably 1,000 hours on it over the years. The CPU benchmark used HPL linpack following a standard-ish Top500-style benchmark, mostly because I think it's fun to see how various single CPU systems compare to historic 'supercomputers' on the official lists. There are different ways to calculate (and benchmark) flops, the way I'm doing it is with this open source project: https://github.com/geerlingguy/top500-benchmark — which can use OpenBLAS, Blis, or ATLAS, and I've tried all three un-tuned. I also worked with some Ampere engineers to run their optimized version for Ampere Altra Max: https://github.com/AmpereComputing/HPL-on-Ampere-Altra On their own server systems with 8 channels of memory they topped out around 1.6 Tflops. There are other tests you can run and get more or

Not super fast but efficiency is okay: https://github.com/geerlingguy/top500-benchmark#results