Java Throughput/Latency & Power Efficiency Tuning For AMD EPYC Turin
With some of these Java benchmarks there wasn't too much difference out of the Java latency and Java throughput optimizations compared to the stock/default performance. This isn't too surprising though since as noted with the EPYC 9575F the default TDP is already at the max cTDP for this OPN. But for those wondering about the power efficiency impact, there is a clear difference. There is also the possibility of even greater power efficiency if using the AMD P-State driver on EPYC Turin as the new default on Linux 6.13+. Unfortunately though the AMD P-State driver isn't yet working on this Supermicro motherboard and thus the ACPI CPUFreq as the existing default on prior kernel versions and for pre-Turin EPYC processors.
For some workloads like the Apache Lucene search engine from DaCapo Benchmark, the Java throughput mode did yield the best performance while the power efficiency mode was indeed showing the best performance-per-Watt on this AMD EPYC 64-core server.
The tuning was worthwhile for performance and power efficiency if desired for the Java Tradebeans benchmark.
Similarly, the Tradesoap benchmark also saw benefit from this Java tuning on 5th Gen AMD EPYC.