Allright fags, let's share what patches and tweaks you use to increase your Linux Desktop performance. Here are mine:
> MuQSS CPU scheduler
> Low latency preemption model
> BFQ I/O scheduler for SSD and HDD
> 1000 hz Timer Frequency
> "Performance" as default CPU governor
> Kernel parameters at grub to disable Spectre and Meltdown patches "pti=off spectre_v2=off l1tf=off nospec_store_bypass_disable no_stf_barrier"
> noatime for ext4 mounts at fstab file for a faster boot
Allright fags, let's share what patches and tweaks you use to increase your Linux Desktop performance. Here are mine:
Jayden Gray
Other urls found in this thread:
Jacob Brown
uninstall dbus
James Wright
>let's share what patches and tweaks you use to increase your Linux Desktop performance
OC'd both my CPU and GPU
> MuQSS CPU scheduler
> Low latency preemption model
> BFQ I/O scheduler for SSD and HDD
> 1000 hz Timer Frequency
> "Performance" as default CPU governor
> Kernel parameters at grub to disable Spectre and Meltdown patches "pti=off spectre_v2=off l1tf=off nospec_store_bypass_disable no_stf_barrier"
> noatime for ext4 mounts at fstab file for a faster boot
that's a lot of alternative medicine placebo bullshit
Nathaniel Perez
Windows 10 is the best patch for linux
Jack Gutierrez
thx for those kernel parameters user, laptop finally doesn't feel bogged down anymore
Cameron Robinson
>hurr i installed a liquorix kernel and turned off spectre/meltdown patches im so 1337
Ayden Powell
This is actually a good thread. Well done OP.
Joshua Cruz
outta here with your bloatnet
Colton Cooper
I dont use a patch just a minimal kernel config but that is crazy to explain here. Besides that I use LILO as the bootloader and JFS. Apparently this filesystem on my laptop made it blazing fast, is like a minimal set of features and maybe that is why works so light.
Robert Barnes
linux-rt
Isaiah Lewis
is there any way to measure the performance gains of any of these?
Joshua Jackson
>> 1000 hz Timer Frequency
You fucking idiot. Higher hz = more overhead. And the scheduler you are using recomends as low a rate as possible. I bet you are some nigger who thinks hes clever just cause he changed a bunch of default settings.
I bet you run EXT4 on your SSD as well.
Wyatt Myers
Choose the appropriate filesystem for your machine.
>JFS
Good read/write performance and crash recovery, reliable for low end computers like laptops and old computers.
>Reiser4
High performance filesystem for use in performance intensive environments like multimedia workstations and gaming rigs.
>EXT4
The newest version of the classic Linux filesystem and the one with most support, excels at a high quantity of files and nested directories. Ideal for FTP and fileservers.
>XFS
Popular for its good handling of large files, its best use is in big, enterprise level databases.
ZFS and BTRFS are still experimental and people complain.
Caleb Allen
Set the niceness of Gnome Shell to -20 and MAYBE it won't lag on a shit computer. If you try it, tell us how it goes.
Josiah Miller
Yeah wondering the same. Else likely placebo esque
Alexander Cooper
I got my information on filesystem from phoronix and a few extra articles.
Mason Jackson
Did you measure the performance boost of each of these, or just blindly follow the kernel option cargo cult from stuff found on random forums?
Carson Young
Not OP but I did some testing on some of these.
>MuQSS CPU scheduler
Placebo nonsense.
>Low latency preemption model
This only mattered a long time ago for realtime audio. Placebo for anything else.
>BFQ I/O scheduler for SSD and HDD
You can find benchmarks on Phoronix on this. AFAIR, the default CFQ generally performans the best.
>"Performance" as default CPU governor
This one truely matters. Not bullshit.
> Kernel parameters at grub to disable Spectre and Meltdown patches "pti=off spectre_v2=off l1tf=off nospec_store_bypass_disable no_stf_barrier"
I did a microbenchmark with a syscall that was 5% slower with mitigations, although this is on an AMD CPU where less of them are enabled.
Camden Price
Link me to the cargo cult forums senpai.
Evan Lewis
imagine being such a brainlet that you can't turn off all Microdick's tracking services.
Jackson Cruz
@70415996
Josiah Walker
Just do a search. Lots of people saying these things should be changed, with little discussion of what it actually does or what the tradeoff is. If it was such a no-brainer, it would be set that way by default.
Some of these like low latency preempt trade throughput for latency. Allowing interruption of kernel code at any time generally lowers performance, but ensures responsiveness which may be desirable on a desktop - but for the most part won't be noticeable unless you're dealing with audio with tiny buffers. Others like Spectre/Meltdown mitigations trade security for a small amount of performance, and the amount depends on what CPU you have and should really be measured before you decide.
Henry Brooks
The higher hz = more frequent kernel wakes, which means lower latency you dumb fuck.
Colton Sanchez
pacman -S linux
Benjamin Brown
Most of this is placebo
Liam Jenkins
>not using full tickless
Wyatt Taylor
> Phoronix benchmark
Stopped reading there. Most of Phoronix benchmarks are stupid unreliable shit. That site is literally Linus Tech Tips of Linux world. None of that benchmark of BFQ show real advantage for DESKTOP tasks of scheduler only it's throughput capabilities. Here is the real benchnark of BFQ and real daily DESKTOP tasks where bfq ownes other schedulers.
Sebastian Turner
What's wrong with ext4 on ssd?
Hudson Perry
Nothing
Nathan Torres
>Most of Phoronix benchmarks are stupid unreliable shit
I agree that many of the benchmarks on there are nonsensical or just flat out testing the wrong thing, but the IO scheduler benchmarks seemed okay and comprehensively compared in various scenarios.
Josiah Harris
muqss gave me ~30 fps boost for CS GO. Source engine strongly rely on CPU even more than on GPU, that's why increase is dramatic. Muqss works best with shittier processors like mine 3rd gen intel i3.
Levi Baker
He could be implying to use F2FS, some people use it for flash memory.