AMD EPYC 7003 "Milan" Linux Benchmarks - Superb Performance

Written by Michael Larabel in Processors on 15 March 2021. Page 4 of 12. 42 Comments

First up are the results of some Chess benchmarks including Crafty, TSCP, Lczero, Stockfish, and Asmfish for some thread-happy tests to kick things off... Going from the top-end EPYC 7742 2P to 7763 2P yielded a 32% generational improvement. When looking at the results compared to the EPYC 7601 2P, keep in mind that original flagship Naples part was a 32-core / 64-thread chip compared to today's flagship 64-core / 128-thread CPUs with Rome and Milan. Meanwhile going from the EPYC 7F52 2P to EPYC 75F3 high frequency SKUs was a 73% improvement. For these chess benchmarks it puts the EPYC 7763 2P at 77% faster than the Xeon Platinum 8280 2P current flagship model in the Cascade Lake line-up.

Next up is a look at the overall code compilation performance based on the results from the timed compilation of the Linux kernel, ImageMagick, LLVM, FFmpeg, and Godot game engine. From the EPYC 7742 2P to 7763 2P was a healthy 14% improvement or from the 7742 2P to 75F3 frequency-optimized SKU was a 19.5% improvement. Compared to the Xeon Platinum 8280 2P is around a 33% advantage for EPYC Milan in code compilation performance.

When looking at various compression tests using 7-Zip, Zstd, and LX4, going from the top-tier Rome to Milan dual socket processors yielded a 28% improvement in performance. Or between the frequency-optimized SKUs, a 43% improvement.

The very significant generational improvements from Rome to Milan kept on delivering. The "creator" tests here included OSPray, Blender, Appleseed, LuxCoreRender, V-RAY, IndigoBench, OCRMyPDF, x264, x265, Kvazaar, dav1d, SVT-AV1, avifenc, libraw, Webp2, JPEG-XL, oneDNN, OIDN, OpenVKL, OpenVINO, Basis Universal, ASTCenc, ETCpak, Godot compilation, eSpeak, Synthmark, and Ngspice. On the top-end is a 24% improvement to the performance and 41% faster than Intel's Xeon Platinum 8280 2P.

Related Articles