Project C v5.9-r1 is released at https://gitlab.com/alfredchen/linux-prjc/-/releases/v5.9-prjc-r1
Please be noticed that you need to pick up below fix if your set your NR_CPUS > 64 in your kernel config.
Thank you very much.ReplyDelete
Interesting, guess I'll be changing my arch config. From the config info:ReplyDelete
Maximum number of CPUs found in arch/x86/Kconfig
depends on: CONFIG_SMP && ! CONFIG_MAXSMP
defined in arch/x86/Kconfig
found in Linux kernels: 2.6.29–2.6.39, 3.0–3.19, 4.0–4.20, 5.0–5.9, 5.9+HEAD
This allows you to specify the maximum number of CPUs which this kernel will support. If CPUMASK_OFFSTACK is enabled, the maximum supported value is 8192, otherwise the maximum value is 512. The minimum value which makes sense is 2. This is purely to save memory: each supported CPU adds about 8KB to the kernel image.
5.9-r1 works not as good as 5.9-r0 for gaming.
The new balancing code seems to introduce more latency for this particular load.
It's better to specify which scheduler(bmq or pds) or both have the latency regression for this gaming particular load. Also provide CPU information too, that will be helpful for the changes in this release.Delete
And, if you can, pls also test which commit cause the regression, so we can have a closer investigation.
It's pds, i5-8600 6core cpu it has no smt.Delete
Which games are we talking about?Delete
What is the difference vs previous version?
What HZ values the kernels have?
Does some benchmark app show differences (Unigine Valley / Heaven / Superposition)?
I can say, that for Doom Etwrnal theres no visible difference, at least on my Ruzen 7 1st gen system.
But, I could try running some benchmark to verify.
It's Quake Champions, very latency sensitive game - I'm running it on wine.ReplyDelete
I'm noticing more input/output latency in 5.9-r1. I'm not sure if an inexperienced person on an low HZ monitor is able to notice the difference.
Kernel is running on 1000HZ
Right, there is no difference in throughput, frame rates are good with 5.9-r1 but there is more latency.
Just notice that my current test machine with 8259U, which share a large 128MB L4 cache. This may lead to a wrong conclusion based on the test results on it. So I may need some time to retest on a CPU with different cache topology.Delete
Thanks for your reporting, I believe you can revert back to r0. I will re-consider best cpu selection in the incoming releases.
Alfred, do you have some ideas how to get actual numbers about this? Ftrace somehow?ReplyDelete