Linux 5.5-rc3 Benchmarks Are Still Pointing To Slips In Performance

[Michael]

Not sure but ...is this related to some new mitigations like JCC ?

Nope, JCC doesn't affect AMD and JCC was just a microcode update. The other recent mitigation work was all in 5.4.

[CochainComplex]

The kernel seems to keep getting worse with every release, and I feel like the worries that the CoC is to blame might be right, and that code standards have dropped.

Things like these make me consider switching to a different kernel, but there's basically nothing that comes close to Linux in basic things such as performance and hardware compatibility, even NT.

This statement makes me wonder if you know the difference between correlation and causality. The kernel got worse in performance since we have uncovered spectre meltdown and all related security flaws....

Would you also relate increase of clogged sewage pipes because you mother has stopped screaming ? ...common

[CochainComplex]

Nope, JCC doesn't affect AMD and JCC was just a microcode update. The other recent mitigation work was all in 5.4.

Ok thank you for resolveing this. I thought there has been some "leftovers" fixed now in 5.5.

[Guest]

This statement makes me wonder if you know the difference between correlation and causality. The kernel got worse in performance since we have uncovered spectre meltdown and all related security flaws....

Would you also relate increase of clogged sewage pipes because you mother has stopped screaming ? ...common

I don't recall any mitigations getting added in this release.

[aufkrawall]

Do you mean 5.5.-rc2 ? If so I also had a huge impact of rc3 vs rc2 ...i thought I screwed up some patches...

Holiday confusion, I actually meant 5.5-rc3.
I tested Shadow of the Tomb Raider in both CPU and GPU limited scenarios, which showed no oddities.
Sure, not very comprehensive testing, but at least it's a game that maxes out CPU threads and GPU too. I think the game's threads are also heavily dependent on each other.

[Guest]

Oh yikes! Some of these regressions are quite significant. I'm a bit surprised, but fairly confident they'll be addressed before final release.

And gosh darn it, I can't test the R9-390 suspend/resume regression now because my son forced me to switch it to my Windows 10 VM so I can play Destiny 2 with him. It works fantastically in the VM, but now I have an ultra crappy GT 710 when running Linux. I sure wish I'd kept the RX 570 I'd bought to replace it, but at the time I kept having random crashes in the VM and didn't realize it was because I was still using the 440fx emulation. So I ended up returning it.

In my defense I'd tried the q35 numerous times over the years but could never get passthrough to work at all, and didn't know how far it had advanced in the last two years. I've switched to it now though, and it's awesome!

Damn, I forgot about that. Looks like I'll have to hold kernel updates until that's fixed. Or, just switch to using the nice Intel iGPU that just works.