Phoronix posted an article comparing 15 different OS versions. This posting doesn’t reproduce the measurements, but instead looks at overall summaries made as well as looks at the benchmarks being used for comparisons.
In the table below is a comparison of individual benchmarks being measured. These comparisons include Windows 10 and the Windows System for Linux. Testing was done on an Intel i9-7980. Graphics performance is in a later article as are BSD operating systems.
Note that one of the comments points out what is done on compress-zstd:
So i did some testing oh how clearlinux got that crazy zstd compression result. Basically, they sort of cheat, they have a patch which always enables multithreading, see https://github.com/clearlinux-pkgs/zstd/blob/master/multi-thread-default.patch. I did some testing on Solus enabling various optimisations e.g. PGO, speed flags (e.g. 03), funroll-loops, etc. which only had a negligible performance increase for the system/compress-zstd benchmark.
|Benchmark||Phoronix observations||My observations||Analysis|
|sqlite||WSL and then Windows are much slower than Linux native. openSuSE tumbleweed is fastest.||Database in one big file with file locking, testing kernel performance.||Analysis|
|fio||Debian slowest, clear Linux fastest for 4KB read/write||fio is flexible with phoronix enabling 8192 different combinations depending on block size, engine, sequential/random, read/write, buffered, direct, etc. Also depends on the storage medium used and latencies in the OS.||Analysis|
|fs-mark||Range of 35%. clearlinux and then debian fastest, Fedora and WSL are slower.||file system testing with varying amounts of files/operations and concurrent threads||Analysis|
|compilebench||WSL pathetically slow. Clearlinux fastests, others behind.||file system testing with large amounts of I/O operations. Response much more important than overall CPU throughput.||Analysis|
|blake2||CentOS fastest, Windows middle of the pack||Single threaded test that completes in less than 0.5 seconds.||Analysis|
|go-benchmark||Clearlinux fastests, WSL slowest on http; json not as big differences with clearlinux fastest; Windows slow on builds; debian slow on garbage collection.||Version of go used makes a big difference. Four somewhat different benchmarks: http higher system time and in parallel; json parallel with lower system time; build mostly sequential; garbage parallel||Analysis|
|cachebench||Windows 10 does well, Ubuntu 18.04, CentOS, WSL lag.||Strides through memory with operations; unfortunately inadequate protection against either prefetching or compiler optimization.||Analysis|
|x264||Windows 10 and Clearlinux highest, WSL low||On_CPU 71%, many voluntary context switches and I/O read input. IPC 1.3||Analysis|
|graphics-magick||Resize only test measured. Windows variants slowest; debian fastest.||Benchmark has five operations; overall On_CPU 30%. A pool of backend processing threads but not always busy. Backend stalls are largest issue.||Analysis|
|stockfish||Close overall with leap highest and sabayon lowest.||On_CPU 100%, IPC of 1.0 with frontend stalls and bad speculation.||Analysis|
|ebizzy||CentOS low, debian very high||Testing kernel memory management with very low IPC and higher backend stalls||Analysis|
|build-linux-kernel||WSL slowest with I/O. Clearlinux fastest.||On_CPU 88%, mostly parallel compiles with a sequential period at end. High frontend stalls. # processes less in subsequent runs so might not do thorough "clean".|
Depends on gcc version - so likely a similar version between distros tested.
|build-php||CentOS and Debian faster with older compiler. WSL slow and clearlinux slow||On_CPU 82% so less parallel times than build-linux-kernel. Frontend stalls high. Many small short-lived processes.||Analysis|
|y-cruncher||Windows slowest, others close with clearlinux slight leader.||On_CPU of 90% with 95% voluntary context switches, some I/O and some thread switching. Backend stalls and overall IPC near 1.0.||Analysis|
|ffmpeg||Clearlinux considerably faster. Otherwise Debian somewhat faster.||Test is On_CPU only 68% much less on AMD with more cores. Workload seems to max out at 2 threads. Otherwise backend stall limited.|
Clearlinux likely doing some things specific to an ill-tuned benchmark.
|tjbench||Debian and clearlinux slightly faster.||Single threaded, high IPC with some branch misses. Runs in 6 seconds.||Analysis|
|pgbench||WSL is low but others are close with Fedora at the highest score.||PostgreSQL database and frontend driver with multiple options. Read keeps cores busier than read/write but light usage overage - at least as much latency issues. Frontend stalls are the largest issues, e.g. icache.||Analysis|
|pybench||OpenSuSE very slow. Others closer with clearlinux as fastest.||Single threaded micro-benchmarks of different python operations. IPC over 2.5 with frontend stalls the largest issue.||Analysis|
|phpbench||OpenSuSE very slow, clearlinux fastest, with factor of 4.5x separating them.||Single threaded, micro-benchmarks of php operations. IPC of 2.76 with ~15% of frontend stalls and ~15% of backend stalls.||Analysis|
|git||WSL and Windows slowest. Linux distributions are close with clearlinux slight leader.||Mostly a test of filesystem operations. single-threaded. IPC 2.07 with high level of branch misses.||Analysis|
|osbench||WSL and windows slow at all the different OS operations tested.||Microbenchmarks, run for 5 seconds and count iterations completed for kernel events. Create processes doesn't cleanly handle fork(2) failures so may not produce a result.||Analysis|
|m-queens||Windows 10 and Debian slightly slower. Others close with OpenSuSE leap fastest.||Benchmark is On_CPU 99%. The IPC is 1.19 with a high level of speculation misses.||Analysis|
|compress-zstd||Clearlinux much faster. debian slowest with sabayon also slower.||Single-threaded benchmark (compress-zstd seems to use multiple threads), with a high level of backend stalls.||Analysis|