NVIDIA GH200 Grace CPU vs. AMD EPYC 9005 Turin CPU Performance

[phoronix]

Phoronix: NVIDIA GH200 Grace CPU vs. AMD EPYC 9005 Turin CPU Performance

With the AMD EPYC 9005 "Turin" testing over the past month since launch I have looked at how well the new EPYC Turin CPUs compete against Intel Xeon, how Turin Dense dominates in performance and power efficiency to AmpereOne at 192 cores, and the generational uplift from EPYC Genoa to Turin at the same core counts, among other Turin performance benchmark tests. Up for comparison today is a look at how the NVIDIA Grace CPU performance within the GH200 Superchip compares to the AMD EPYC Turin processors.

https://www.phoronix.com/review/nvidia-grace-epyc-turin

[schmidtbag]

Seems like pretty healthy competition - you can either opt for raw performance or good efficiency. Contrast this to most other options where there isn't a clear advantage.

[r1348]

Pretty much aligned with other ARM server solutions: more efficient, but not reaching the same performance peak of x86.

[isaacx123]

Pretty much aligned with other ARM server solutions: more efficient, but not reaching the same performance peak of x86.

Is it actually more efficient though?
Not according to the geomean/W.

[bug77]

These benchmarks are begging for a follow up. Grace either dominates or seriously lags behind in most tests. Are we looking at some kind of scheduler inefficiency? Is the hardware tuned towards certain workloads?

[creative]

Strange and interesting.

[varikonniemi]

Is it actually more efficient though?
Not according to the geomean/W.

how so? The performance was half, and power consumption less than half of the premiere amd cpu

edit: oh, i thought the vertical black line is the average but apparently not when looking at the numbers on the left. What does it denote then?

[bug77]

how so? The performance was half, and power consumption less than half of the premiere amd cpu

edit: oh, i thought the vertical black line is the average but apparently not when looking at the numbers on the left. What does it denote then?

If you do the math, Grace ends up between the 9655 and 9755.
However, considering the uneven performance, that means it will crush them in perf/W in some workloads and it will lose badly in others.

[Hell.Hand]

It seems to me that some tests aren't scaling well with EPYC's massive number of cores. With proper optimization, Turin should win in 90% of the scenarios, both in efficiency and performance.

[coder]

Michael , thanks for the benchmarks!

BTW, does --enable-multiarch provide functionality similar to -march=native? I'm just curious whether the Grace CPU gets to use its SVE2 capability, which is specific to ARMv9-A.