XanMod, Liquorix Kernels Offer Some Advantages On AMD Ryzen 5 Notebook

[CochainComplex]

I'm going to share my configs too. One with amdgpu built in for anyone that has Polaris that wants, and one that just builds it as a module like normal. I find building it in is preferred for smoother bootup and such.

But I'm telling you, if everything is configured right, it should run flawless. Should be no drawbacks or nuances

I was thinking about sharing too. but need to finallyze the scripts. I dont know how they work outside of my system...and I need some more experience for "compiling" patchsets.

[schmidtbag]

Seems Liquorix does a better job optimizing the kernel than AMD themselves.

[ms178]

Some guy was sweating me on here yesterday saying tweaking the kernel is pointless, talking about "microbenchmarks" when it's all relative, bud. Same exact machine, different kernel. You're tweaking the brains of the OS and you don't find that of interest or worthwhile?

Just benched my customized vanilla 5.12.19 + Speculative page faults patchset + march=native and some other fancy flags against a Ubuntu standard Kernel: 86 fps vs. 39 fps, with the same custom Mesa and libdrm for each setup, the game is Company of Heroes 2. People who claim that there was no point in all of this should try it out and bench their favorite application or game first before making such statements. In my case the outcome is more than worth the effort.

[CochainComplex]

Just benched my customized vanilla 5.12.19 + Speculative page faults patchset + march=native and some other fancy flags against a Ubuntu standard Kernel: 86 fps vs. 39 fps, with the same custom Mesa and libdrm for each setup, the game is Company of Heroes 2. People who claim that there was no point in all of this should try it out and bench their favorite application or game first before making such statements. In my case the outcome is more than worth the effort.

absolutely! I can recommend the Clear Linux std. "-O3 -falign-functions=32 -fno-math-errno -fno-semantic-interposition -fno-trapping-math"

just figured out that they have changed the flags for mesa "$CFLAGS -Ofast -falign-functions=32 -fno-lto -fno-semantic-interposition -mprefer-vector-width=256 "

however it is really worth to crawl through this pkg repo and search for the *.spec files. just copy past the flags and test it.

https://github.com/clearlinux-pkgs

[perpetually high]

absolutely! I can recommend the Clear Linux std. "-O3 -falign-functions=32 -fno-math-errno -fno-semantic-interposition -fno-trapping-math"

just figured out that they have changed the flags for mesa "$CFLAGS -Ofast -falign-functions=32 -fno-lto -fno-semantic-interposition -mprefer-vector-width=256 "

however it is really worth to crawl through this pkg repo and search for the *.spec files. just copy past the flags and test it.

https://github.com/clearlinux-pkgs

Agreed. I'm finishing up some stuff, but I'm going to share my CFLAGS, sysctl.conf, and my GRUB. I think that covers most things. Will likely just throw it on my personal github (unfortunately) for easy sharing of the script and patches and so I can update it as time goes on

[HD7950]

My choice is kernel-pf: https://gitlab.com/post-factum/pf-kernel/-/wikis/README

Is anyone else using it?

[CochainComplex]

My choice is kernel-pf: https://gitlab.com/post-factum/pf-kernel/-/wikis/README

Is anyone else using it?

just flew over the ingredients list

haven't seen clear linux patches in the pf kernel.

By the patches used for pf one could conclude in mathematical terms that the pf patchset is a subset of the xanmod patchset.

https://xanmod.org/

[Linuxxx]

I totally understand Michael's reasoning for testing only the defaults, but when all of the upstream kernels are defaulting to "schedutil" while all of the custom ones default to the "performance" CPU governor, that still renders these benchmarks mostly useless, unfortunately!

Well, if anything, this round of testing actually shows how far 'schedutil' has advanced on AMD hardware, so that's definitely a plus for the Linux kernel as a whole!

[Vistaus]

just flew over the ingredients list

haven't seen clear linux patches in the pf kernel.

By the patches used for pf one could conclude in mathematical terms that the pf patchset is a subset of the xanmod patchset.

https://xanmod.org/

…except that pf is older, so it's actually the other way around.

[CochainComplex]

…except that pf is older, so it's actually the other way around.

ok this might be true but it does not contradict my point If you use xanmod you will have at least all what the pf kernel is offering thats why in math terms it is a subset of xanmod. But sure it might be possible that xanmod was innitially a copy of the pf kernel - this i dont know.