Help wanted: benchmarking multi-threaded CPU performance

Nvidia Grace superchip (144 cores):

~/grace$ julia --threads=auto cpu.jl
...
[39.3, 397.4, 3.4, 12.2, 206.5, 383.5, 392.9, 341.4, 261.2]

And

:~/grace$ julia --threads=72 cpu.jl
...
[219.6, 224.1, 219.1, 225.4, 202.1, 218.2, 209.1, 216.1, 188.0]
4 Likes