The AVX-512 Performance Advantage With AMD EPYC Bergamo

Across all the AVX-512 capable workloads tested, on average the performance effectively doubled. Of course, some workloads see more dramatic performance improvements than others... Those interested can see all the individual raw results here.

Making use of AVX-512 overall didn't cause any significant difference in the CPU peak frequency being achieved with the EPYC 9754. On average the peak frequency throughout all the dozens of tests conducted was at around 2.9GHz.

With the Bergamo AVX-512 testing we did see slightly higher power consumption compared to looking at the AVX-512 performance with the Ryzen 7000 series or EPYC 9004 Genoa(X) parts. Presumably some of that higher AVX-512 power difference is due to simply having more cores and any other alterations as part of the Zen 4C design. AVX-512 caused the CPU power consumption to increase by about 50 Watts or 20 Watts higher on the peak CPU power consumption observed. But for many tests the CPU power consumption was similar between runs. The power increase is also worthwhile when considering the much larger increases in performance observed for many workloads.

With the slightly higher CPU power consumption the EPYC 9754 core temperature came put higher with AVX-512 enabled. But with a 51 degrees average and 74 degrees peak it's still efficient and unlike the early days of AVX-512 on Xeon processors where it meant excessive power consumption and thermal throttling.

So, long story short, the AVX-512 implementation with the Zen 4C cores on AMD EPYC Bergamo still prove very beneficial and great to see Advanced Vector Extensions 512 with this denser, cloud-focused server processor. In case you missed it from last week are also many other Bergamo benchmarks on Phoronix.