Can numactl be used to change the memory allocation strategy program by program?Seeing your post, I also re-ran tinymembench, and found a similar regression—in my case memset going from 13+ GB/sec to 9.3 GB/sec. See my full test results here.Ran some tests and things seem to be working well overall. Geekbench 6 is indeed faster than ever before. So is Google Octane v2. Passmark performs extremely good as well, except for the "Memory Write" sub test, which regresses from ~11500 MB/s to 8000 something. Sysbench memory read performance is also better than ever (very nice latency, both average and max), while the write test corroborates Passmark's result, showing a severe regression in write performance.
@dom - Here's how I'm running `tinymembench` for a reproducer:The lower synthetic speeds don't result in other benchmarks running any slower though. It's just sad that not every number goes up and to the rightCode:
git clone https://github.com/rojaster/tinymembench.gitcd tinymembench && make./tinymembench
Statistics: Posted by ejolson — Mon Dec 02, 2024 11:22 pm