Benchmarking Linux 5.5 vs. Linux 5.6-rc1 On A Few Systems So Far
Linux 5.5 vs. 5.6-rc1 were benchmarked using the reference binaries from the Ubuntu Mainline Kernel PPA. For this quick article are results from a Threadripper 3970X, AMD EPYC 7742 2P, and Intel Xeon Platinum 8280 2P in distinctly different configurations in seeing if there is any widespread trends as of 5.6-rc1 for these high-end systems.
On the Threadripper 3970X system with Corsair Force MP600 1TB PCIe 4.0 NVMe storage, most of the differences between the kernels came down to I/O.
FIO saw several cases where Linux 5.5 was much faster than 5.6-rc1 but there were also a number of FIO test cases where Linux 5.6 was doing better. Most of the other differences on this 3970X system came down to within a few percent differences. All the results in full can be found via this OpenBenchmarking.org result file.
Next up was the dual EPYC 7742 Rome server.
It saw a few nice wins when it came to network/socket performance, memory allocations, and like Threadripper did also see a number of different outcomes with the FIO I/O performance. All of that EPYC Rome data via this OpenBenchmarking.org result file.
For the last of this quick one-page article was the dual Xeon Platinum 8280 performance.
The Intel two-socket Cascade Lake server did see some FIO results taking up the leaderboard but did see some improvements out of the synthetic Hackbench and Stress-NG tests. There were also some minor losses. To some surprise, in the compilation tests run that make use of GNU Make there didn't end up being much of a difference in compilation speed at least with these high-end systems tested given that Linux 5.6 has the pipe optimization that should help out Make. In any case, more of the Xeon Platinum 8280 benchmarks via this OpenBenchmarking.org result file while more tests on different systems are still ongoing.
3 Comments
