AMD's Epic Performance Gains From The Original EPYC 7601 To EPYC 9755 / EPYC 9965
Compiling the full Linux x86_64 kernel was ~4.6x faster on the EPYC Turin processors than the original EPYC Naples. Though this isn't even an ideal workload given the compiler scalability not scaling well to 256~384 threads when compiling a single large codebase. Those running a CI/CD farm with multiple concurrent builds happening are likely to see even greater build time improvements.
Compiling Node.js on the EPYC 9965 was 7.4x faster than using the eight year old EPYC 7601 server. Plus far better power efficiency too for both the EPYC 9755 and EPYC 9965 processors compared to the EPYC 7601, even with the 500 Watt TDP rating.
Turing to Blender and other creator workloads is where the EPYC Turin processors can really show how far AMD EPYC performance has evolved since the Zen 1 days. The well known BMW scene with Blender took 75 seconds to render on the EPYC 7601 server while just 8~9 seconds with today's high-end Turin processors.
More than 9x the performance for some of the scenes rendered with the Blender 3D modeling software.
With workloads able to leverage AVX-512 on Zen 5 (and Zen 4) server processors, the gains become really mind-boggling to look at the AMD server performance gains over the past eight years. With Intel's OSPRay ray-tracing engine is more than 21x the performance from EPYC 7601 to EPYC 9965.