The AVX-512 Performance Advantage With AMD EPYC Bergamo

Written by Michael Larabel in Processors on 26 July 2023 at 03:30 PM EDT. Page 2 of 5. 85 Comments.
miniBUDE benchmark with settings of Implementation: OpenMP, Input Deck: BM1. AVX512 On was the fastest.
miniBUDE benchmark with settings of Implementation: OpenMP, Input Deck: BM1. AVX512 On was the fastest.
miniBUDE benchmark with settings of Implementation: OpenMP, Input Deck: BM2. AVX512 On was the fastest.
miniBUDE benchmark with settings of Implementation: OpenMP, Input Deck: BM2. AVX512 On was the fastest.
libxsmm benchmark with settings of M N K: 128. AVX512 On was the fastest.
libxsmm benchmark with settings of M N K: 256. AVX512 On was the fastest.

Not that I was expecting anything dramatically different, but the AVX-512 presence on the 128-core EPYC 9754 showed to certainly be of help in the relevant HPC benchmarks.

libxsmm benchmark with settings of M N K: 256. AVX512 On was the fastest.
libxsmm benchmark with settings of M N K: 256. AVX512 On was the fastest.

AVX-512 with Zen 4C proved efficient with the libxsmm test for example showing no difference in the CPU power consumption when it was leveraged.

libxsmm benchmark with settings of M N K: 256. AVX512 On was the fastest.
libxsmm benchmark with settings of M N K: 256. AVX512 On was the fastest.

When AVX-512 was being utilized in libxsmm there was also no measurable difference in the CPU peak frequency being achieved nor the core temperature.

Embree benchmark with settings of Binary: Pathtracer ISPC, Model: Crown. AVX512 On was the fastest.
Embree benchmark with settings of Binary: Pathtracer ISPC, Model: Asian Dragon. AVX512 On was the fastest.
Embree benchmark with settings of Binary: Pathtracer ISPC, Model: Asian Dragon Obj. AVX512 On was the fastest.

Even with Intel's own (excellent) open-source creator software like Embree the AVX-512 with Bergamo proved to be advantageous.


Related Articles