Last week with my ongoing testing of the in-development Linux 7.0 kernel I found nice performance improvements for PostgreSQL and other workloads when testing on a 128-core AMD EPYC 9755 “Turin” server. Curious if those wins were due to optimizations focused on better scalability with today’s “big” servers, I also ran some comparison Linux 7.0 benchmarks on the smaller AMD EPYC 4005 class servers too. Some nice wins carried over.
Using an AMD EPYC 4585PX server for entry-level servers I ran benchmarks of Linux 6.19 stable against Linux 7.0 Git as of 24 February.
Same software/hardware and same kernel configuration (besides defaults for new v7.0 kernel options) in just looking at the impact of the new kernel on the AMD EPYC 4585PX 16-core server processor running within a Supermicro AS-3015A-I H13SAE-MF server.
A varity of benchmarks were then run to see the impact of Linux 7.0 on this AMD Zen 5 16-core system.
The workloads with statistically significant changes on Linux 7.0 ended up being databases, similar to what I found with the big EPYC 9005 Turin server testing too. Some nice wins for PostgreSQL as well as MariaDB and Redis and CockroachDB. Outside of database workloads were some smaller wins for OpenVINO and CloverLeaf.
Outside of those workloads the performance was flat overall for this Linux 6.19 vs. Linux 7.0 performance benchmarking with no other big takeaways. At least no measurable regressions yet observed on Linux 7.0 in my benchmarking thus far.
These database wins from small to big AMD EPYC servers at least is nice to see given that Linux 7.0 is set to be the default kernel of the upcoming Ubuntu 26.04 LTS release.
