NVIDIA GH200 CPU Performance Benchmarks Against AMD EPYC Zen 4 & Intel Xeon Emerald Rapids

With the Incompact3D HPC benchmarks that crave as much memory bandwidth as possible, the GH200 was competing with the other 1P AMD EPYC Genoa processors and ahead of the Xeon Platinum 8592+ Emerald Rapids processor. Again, huge uplift over the current Ampere Altra Max top-end processor -- I am still waiting on any AmpereOne server platform to see how that compares.

With the LULESH hydrodynamics software the GH200 was delivering CPU performance comparable to the EPYC 9554 Genoa and EPYC 9754 Bergamo processors in single socket configurations.

Some workloads aren't too well optimized for AArch64 but over the past few years there has been much progress particularly by prominent open-source projects in better embracing ARM/AArch64 optimizations as more hardware becomes available and more performant Raspberry Pi single board computers and other options become available for more independent developers getting their hands on ARM hardware.

With the OpenMP-threaded GraphicsMagick imaging software the GH200 was striking near the top of the stack.