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

It shouldn't come as much of a surprise given the AVX-512 comparisons I previously carried out with Ryzen 7000 series desktop CPUs, but with workloads supporting AVX-512 it can make a huge difference for 4th Gen EPYC performance... AI, of course, being one of the areas of much potential with AVX-512. Even with the Zen 4 "double pumping" approach, the impact of AVX-512 with the flagship EPYC 9654 processors is huge.

The power consumption for the two EPYC 9654 processors was similar whether or not AVX-512 was disabled. In fact, for AI Benchmark and others in some cases the power draw was slightly lower when AVX-512 was engaged -- quite the opposite for those that recall the Skylake-X AVX-512 days.

So the performance-per-Watt is equally tantalizing with AVX-512.

With other AI workloads like Neural Magic's DeepSparse, there were also significant advantages from employing AVX-512 on Genoa though by not as wide of margins as with AI-Benchmark.

While DeepSparse didn't see quite as big of an improvement as AI-Benchmark and some of the other workloads to be shown, it was still a very sizable benefit and without any negative impact to the CPU clock frequencies or power consumption.