AMD EPYC 9655 Benchmarks Show The Terrific Generational Gains With 5th Gen EPYC
For code compilation workloads of large codebases to help with the CPU scaling across multiple jobs, the EPYC 9655 shows off the significant uplift possible going from 4th Gen EPYC to 5th Gen EPYC at the same 96-core count. Between the EPYC 9655 and EPYC 9755 processors there isn't much difference in these runs and similarly between 1P and 2P due to limited scaling at the very high core counts for compiling a single codebase/project. Those carrying out multiple builds in parallel for CI/CD type servers will find greater scalability with the EPYC 9755 model. In any case here the main takeaway in going from the EPYC 9654 to EPYC 9655 was a ~1.47x improvement in the build speed at the same core count.
Making the faster build times even more impressive is that the single socket CPU power going from the EPYC 9654 to EPYC 9655 went up by just 6 Watts on average or 30 Watts at peak, running comfortably still below the 400 Watt default TDP in this particular test. Compiling the Gem5 codebase required ~32k Joules with the EPYC 9654 and a similar amount with Intel Xeon Granite Rapids but only around ~23k Joules with the EPYC 9655 Turin processor.
With the John The Ripper crypto program, the EPYC 9655 was around 1.42x faster than the EPYC 9654. The EPYC 9655 96-core processor was performing similarly in this OpenMP-threaded application to the 128-core Xeon 6980P Granite Rapids processor.
AMD EPYC Turin really stands out with delivering 1.3x the performance-per-Watt of the Genoa comparison point.
With John The Ripper's WPA PSK benchmark, the EPYC 9655 was performing much stronger than the Xeon 6980P in both single and dual socket configurations. Showing the EPYC 9654 to EPYC 9655 uplift really helps show the Zen 5 strengths beyond the 5th Gen EPYC CPU models with higher core counts too.