AMD 4th Gen EPYC 9654 "Genoa" AVX-512 Performance Analysis

Besides looking at the AVX-512 performance in a number of individual Intel oneAPI libraries/components, when combining it all into the rendering performance for the OSPRay Studio interactive visualization application the AVX-512 results were equally compelling. Zen 4's AVX-512 was delivering very meaningful performance uplift, sometimes even at lower power consumption than without AVX-512, and no thermal throttling to worry about.

The Genoa AVX-512 performance was also looking great for Intel's oneDNN neural network library with the exception of one recurrent neural network inference test case where the AVX-512 usage happened to regress the performance, but that may also be due to the software itself.

Alibaba's Mobile Neural Network (MNN) software was also seeing nice gains with AVX-512 left enabled but with just one regression as well when using their Inception-v3 model.