CachyOS Performance Optimization Guide

For: Intel 8th–10th gen Core i5/i7, NVIDIA Turing/Ampere, 16GB RAM, CachyOS with Wayland.
Kernel: linux-cachyos, PREEMPT, 1000Hz, EEVDF default, sched_ext enabled.

Every section explains what CachyOS gives you out of the box, what to add, and the tradeoffs. All config files are at the bottom: copy, paste, done.

Contents

1. What CachyOS already does
2. Before you begin
3. Boot parameters
4. CPU tuning
5. sched_ext: swappable schedulers
6. Memory & swap
7. Storage & filesystems
8. GPU: NVIDIA
9. GPU overclocking (LACT)
10. Gaming pipeline
11. Audio: PipeWire
12. Network
13. System limits & sysctl
14. Services to disable
15. Desktop & Wayland
16. Minimum viable tuning (80/20)
17. Verification checklist
18. System recovery (if you break it)
19. Configuration file reference

contents

1. What CachyOS already does

CachyOS ships with aggressive defaults. Know what's already tuned so you don't break it or do it twice.

Kernel

The flagship linux-cachyos package (what most installs run) includes PREEMPT (full preemption), 1000Hz tick rate, Clang ThinLTO, x86-64-v3/v4 targeting, and NTSYNC for Wine/Proton. All of this is there from install.

Sysctl defaults

vm.swappiness = 100
vm.vfs_cache_pressure = 50
vm.dirty_bytes = 268435456
vm.dirty_background_bytes = 67108864
vm.dirty_writeback_centisecs = 1500
vm.page-cluster = 0
kernel.nmi_watchdog = 0
kernel.split_lock_mitigate = 0
kernel.printk = 3 3 3 3
kernel.kptr_restrict = 2
net.core.netdev_max_backlog = 4096
fs.file-max = 2097152

ZRAM

zstd compression, RAM-sized, swap priority 100. A udev rule sets vm.swappiness=150 at activation. High swappiness keeps file cache in RAM and pushes cold anonymous pages into compressed ZRAM instead.

Transparent Hugepages

always + defer+madvise defrag + max_ptes_none=409. Benefits tcmalloc apps (Chromium, Proton games) without the historical memory bloat.

IO Schedulers

bfq for HDDs, mq-deadline for SATA SSDs, kyber for NVMe.

NVIDIA module options

NVreg_UsePageAttributeTable=1, NVreg_InitializeSystemMemoryAllocations=0, NVreg_DynamicPowerManagement=0x02. CachyOS sets these for you.

Systemd

Other defaults

rtkit + audio group rtprio 99. SATA ALPM = max_performance. hdparm -B 254 -S 0 on HDDs. iTCO_wdt + sp5100_tco blacklisted. NTP via Cloudflare + Google. DNS via systemd-resolved.

contents

2. Before you begin

Rule #1: Don't fuck up your system without a net. CachyOS defaults to BTRFS. Use it:

sudo btrfs subvolume snapshot / /@pre-optimization

Rule #2: Test one change at a time. Rebooting after applying the whole list is a recipe for not knowing what broke.

contents

3. Boot parameters

CachyOS uses Limine. Parameters go in /etc/default/limine and apply with sudo limine-mkinitcpio. If you use GRUB or systemd-boot the parameters are identical; only where you put them differs.

Recommended command line

quiet mitigations=off nowatchdog nmi_watchdog=0 nvidia_drm.modeset=1
tsc=reliable clocksource=tsc intel_pstate=active preempt=full
split_lock_detect=off pcie_aspm=performance intel_idle.max_cstate=1
transparent_hugepage=madvise splash rw

Kernel update survival guide: After a pacman -Syu, check your NVIDIA params and scheduler availability:

cat /proc/driver/nvidia/params | grep -E "UsePageAttributeTable|InitializeSystemMemoryAllocations"
cat /sys/block/nvme0n1/queue/scheduler  # if ADIOS is gone, revert to kyber

# Edit and apply
sudo nano /etc/default/limine
sudo limine-mkinitcpio
sudo reboot

contents

4. CPU tuning

Lock the governor to performance using three mechanisms; each covers gaps in the others.

Step 1: cpupower service

sudo systemctl enable --now cpupower
echo 'governor="performance"' | sudo tee /etc/cpupower.conf

Step 2: tmpfiles.d: these run at early boot, before any desktop session. A power manager daemon can't override what was set before it started.

# /etc/tmpfiles.d/force-performance.conf
w /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor - - - - performance
# /etc/tmpfiles.d/force-epp-performance.conf

w /sys/devices/system/cpu/cpu*/cpufreq/energy_performance_preference - - - - performance

Verify:

cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
# → performance
cat /sys/devices/system/cpu/cpu0/cpufreq/energy_performance_preference
# → performance

contents

5. sched_ext: swappable schedulers

linux-cachyos ships EEVDF as its default scheduler, not BORE. See the correction in §1. EEVDF vs BORE matters less than what actually sets CachyOS apart here: CONFIG_SCHED_CLASS_EXT=y is built into the default kernel, so sched_ext (SCX) is available from install. SCX schedulers are CPU schedulers implemented as loadable BPF programs. You can swap the whole scheduling algorithm at runtime. No reboot, no separate kernel, and you're back on EEVDF the moment you stop one.

Install

sudo pacman -S scx-scheds scx-tools
# scx_loader (background service, D-Bus API) and scxctl (CLI client) come with scx-tools

If you want the newest experimental schedulers before they land in the stable branch, use scx-scheds-git scx-tools-git instead.

Launching one

# One-off test. Ctrl+C stops it and you're back on EEVDF.
sudo scx_bpfland

# Managed via scxctl, which talks to the scx_loader D-Bus service
sudo systemctl enable --now scx_loader
scxctl start --sched bpfland --mode gaming
scxctl get                     # what's running right now
scxctl stop                    # back to EEVDF

CachyOS Kernel Manager and the standalone SCX Manager GUI both wrap the same scx_loader service if you'd rather click through it than type it.

Which scheduler for this hardware

The i5-9400F is a monolithic 6-core, 6-thread die. No SMT, no heterogeneous P/E cores. That rules out most of the SMT-contention and hybrid-core-aware flags other schedulers expose, so the choice really comes down to raw interactivity under load.

What actually won on this rig

Numbers below are from a real CS2 benchmark run on this exact machine (COSMIC/Wayland, i5-9400F, GTX 1650 at the 75W/1860MHz lock from §9), launch options LOW_LATENCY_LAYER=1 LOW_LATENCY_LAYER_REFLEX=1 SDL_VIDEODRIVER=wayland gamemoderun %command% -sdlaudiodriver pulse -novid, both schedulers on their low-latency/esports profile:

CS2: scx_cake vs scx_lavd (lowlatency, 2 runs)
           Avg FPS   P1 low   P99 (ms)
cake r1      208.9    116.7      8.57
cake r2      208.1    109.1      9.16
cake avg     208.5    112.9      8.87
lavd r1      208.6    108.2      9.24
lavd r2      200.4     91.2     10.96
lavd avg     204.5     99.7     10.10

scx_cake beat scx_lavd on this box, mainly on consistency: cake's 1% low stayed within about 8ms of itself across both runs, lavd's second run dropped a spike that cost it roughly 13ms of 1% low margin against cake's average. That's one bad run out of two though, not a verdict. Worth another 2 to 3 passes before writing lavd off entirely, especially since it's tuned around gaming workloads on paper and might just need a different flag set on a 6-core chip with no SMT than it gets on the hardware it's usually benchmarked on. scx_bpfland and scx_cake were the best performers overall on this setup. Your mileage may genuinely vary since SCX performance is workload- and CPU-topology-dependent. Run your own comparison with schbench/cachyos-benchmarker and the games you actually play before locking anything in.

Autostart on boot

sudo mkdir -p /etc/scx_loader
sudo cp /usr/share/scx_loader/config.toml /etc/scx_loader/config.toml

# /etc/scx_loader/config.toml
default_sched = "scx_cake"
default_mode = "Gaming"

sudo systemctl enable --now scx_loader.service

Swap to bpfland for testing with scxctl switch --sched bpfland --mode gaming, no need to edit the config file for a quick comparison.

ananicy-cpp compatibility

§10 tells you to disable ananicy-cpp because it fights Gamemode's renice values. That's still correct and has nothing to do with sched_ext. Separately, ananicy-cpp used to cause stalls when combined with early SCX schedulers. On current scx-scheds releases the two are generally fine together. If you re-enable ananicy-cpp for other reasons and see stalls, disabling it is still the first thing to try.

Benchmarking before you commit

sudo pacman -S schbench cachyos-benchmarker
schbench -m 2 -t 8 -r 60          # wakeup and request latency percentiles
cachyos-benchmarker ~/bench/      # full suite: compiles, encodes, hashes, renders, per-scheduler

Run the same test across EEVDF (default, nothing loaded), BORE if you install that variant, and 2 to 3 SCX schedulers before picking one permanently. A scheduler that wins on paper for "gaming" can still lose on your specific combination of a 6-core no-SMT budget CPU and one GPU doing all the rendering work. That's exactly what the cake vs lavd numbers above show: run it yourself, don't take a table off a forum post as gospel, including this one.

contents

6. Memory & swap

ZRAM: consider lz4 over zstd

CachyOS defaults to zstd. On a 6-core CPU without hyperthreading, zstd compression eats measurable CPU during memory pressure. Switch to lz4 if you have fewer than 8 threads or notice CPU spikes under load.

# /etc/systemd/zram-generator.conf.d/override.conf
[zram0]
compression-algorithm = lz4

sudo systemctl restart systemd-zram-setup@zram0

Dirty page writeback

Linux defaults allow 20% of RAM (~3.2GB on 16GB) to become dirty before throttling writes. When you download or extract a large file, everything buffers in page cache, then flushes all at once. The resulting I/O storm starves the compositor. You see stutter.

# In /etc/sysctl.d/99-latency.conf:
vm.dirty_background_ratio = 3
vm.dirty_ratio = 8

contents

7. Storage & filesystems

XFS mount options

SSD root:

UUID=... / xfs defaults,lazytime,noatime,inode64,logbsize=256k,noquota 0 1

HDD bulk storage:

UUID=... /mount/point xfs defaults,noatime,nofail,allocsize=64m 0 0

tmpfs on /tmp

tmpfs /tmp tmpfs defaults,noatime,mode=1777 0 0

Puts /tmp in RAM. Reduces SSD writes, speeds up temp file operations. Negligible memory cost on 16GB.

F2FS for secondary SSDs

If you have a second SSD for downloads or temp data, F2FS excels at sequential writes. Ideal for drives where data integrity isn't critical.

sudo mkfs.f2fs -f /dev/sdX  # destructive, wipes the drive

IO scheduler: switch to ADIOS

CachyOS defaults: mq-deadline (SSD), kyber (NVMe), bfq (HDD). ADIOS (Adaptive Disk I/O Scheduler, a CachyOS kernel feature) balances throughput and latency better for single-user desktops:

# /etc/udev/rules.d/60-ioschedulers.rules
# HDD: BFQ

ACTION=="add|change", KERNEL=="sd[a-z]*", ATTR{queue/rotational}=="1", 
ATTR{queue/scheduler}="bfq"

# SSD: ADIOS

ACTION=="add|change", KERNEL=="sd[a-z]*|mmcblk[0-9]*", ATTR{queue/rotational}=="0", 
ATTR{queue/scheduler}="adios"

# NVMe: ADIOS

ACTION=="add|change", KERNEL=="nvme[0-9]*", ATTR{queue/rotational}=="0", 
ATTR{queue/scheduler}="adios"

sudo udevadm control --reload-rules && sudo udevadm trigger

NVMe APST (optional): Some NVMe drives aggressively enter power-saving states. Disable it:

echo 0 | sudo tee /sys/class/nvme/nvme0/device/power/control

contents

8. GPU: NVIDIA

Persistence daemon

sudo systemctl enable --now nvidia-persistenced

Keeps driver loaded with no apps using GPU. Kills the ~1–3s init delay on first launch after boot.

Full RGB over HDMI

NVIDIA defaults to Limited Range (16–235) over HDMI, causing black crush and washed out colors.

# /etc/X11/xorg.conf.d/20-nvidia-full-rgb.conf
Section "Device"
    Identifier "NVIDIA"
    Driver "nvidia"
    Option "ColorSpace" "RGB"
    Option "ColorRange" "Full"
EndSection

Note: If nvidia-settings daemon is running, it might override this. Kill it if needed: sudo pkill nvidia-settings.

Environment variables

Put these in /etc/environment; they apply system-wide:

__GL_THREADED_OPTIMIZATION=1
__GL_SHADER_DISK_CACHE=1
__GL_SHADER_DISK_CACHE_SKIP_CLEANUP=1
__GL_VRR_ALLOWED=0
__GL_SYNC_TO_VBLANK=0
__GL_MaxFramesAllowed=1
VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/nvidia_icd.json
WINEDEBUG=-all

Shader Cache Size

__GL_SHADER_DISK_CACHE=1 enables the cache, but without a size limit, the driver uses its default cap of 1 GB (increased from 128 MB in driver 460). Modern titles (especially DXVK/VKD3D) can generate 2–4 GB of shaders across your library. When the cache exceeds the limit, the driver purges older entries, forcing recompilation on next launch, causing stutter and longer load times.

Set a global size limit to prevent eviction:

mkdir -p ~/.config/environment.d
echo "__GL_SHADER_DISK_CACHE_SIZE=12000000000" > ~/.config/environment.d/shader_cache.conf

Log out and back in, or reboot. Verify with:

env | grep __GL_SHADER_DISK_CACHE_SIZE

The GTX 1650's 4 GB VRAM bounds the shader permutation space. Beyond 4 GB disk cache, you're caching permutations for games you haven't launched in months without improving current-session hit rates.

Optional: clear legacy caches (old .nv path still uses 128 MB default):

rm -rf ~/.nv/GLCache ~/.cache/nvidia/GLCache

The driver will rebuild under the new XDG path (~/.cache/nvidia/GLCache) with the 4 GB cap applied.

NVIDIA Wayland reality check

nvidia_drm.modeset=1 is necessary but not sufficient. Requirements:

• Driver ≥ 550.xx (nvidia-smi --query-gpu=driver_version --format=csv,noheader)
• Compositor with explicit sync (KDE Plasma ≥ 6.1, GNOME ≥ 46, COSMIC latest)

If you get black screens or flickering, switch to X11.

contents

9. GPU overclocking (LACT)

LACT controls NVIDIA GPUs through NVML. GUI for overclocking, fan curves, monitoring.

sudo systemctl enable --now lactd

The clock lock strategy (not "undervolt")

Turing and later NVIDIA GPUs don't support traditional Linux undervolting. Instead: lock GPU clock below stock boost and add a positive P8 voltage offset. The GPU runs at higher voltage than it naturally would at that clock, with no power throttling and cooler running than stock boost.

Fan curve

Quiet at idle-moderate, aggressive past 70°C. Adjust to your card's cooler. Stay under 75°C during extended loads.

Stability: Run a demanding game for 2–3 hours. Reduce VRAM offset by 50MHz if you see artifacts, crashes, or driver errors. Test a full 8+ hour session before calling it stable.

contents

10. Gaming pipeline

Gamemode

Gamemode applies optimizations when you launch with gamemoderun %command%.

# /etc/gamemode.ini
[general]
renice = 20
ioprio = 0
[gpu]
apply = 1
nv_powermizer_mode = 1

[custom]
start = nvidia-smi -pm 1
end = nvidia-smi -pm 0

Gamemode group

# /etc/security/limits.d/10-gamemode.conf
@gamemode - nice -10

sudo usermod -aG gamemode $USER

Steam launch options

Texture quality overrides

Replicates NVIDIA Control Panel "High Quality" filtering. NVIDIA Settings on Linux doesn't expose per-game control.

# DXVK (DirectX 9/10/11 via Proton)
DXVK_ANISO=16 DXVK_LODBIAS=-0.5 PROTON_ENABLE_WAYLAND=1 gamemoderun %command%
# Native OpenGL
__GL_LOG_ANISO=16 __GL_TEXTURE_LOD_BIAS=-0.5 SDL_VIDEODRIVER=wayland gamemoderun %command%

Gamescope: probably skip it

Gamescope adds compositing latency. On a native Wayland compositor with fullscreen bypass (COSMIC, KWin, Mutter), you don't need it. Only use it for FSR upscaling, HDR tonemapping, or nested sessions.

Proton

sudo pacman -S proton-cachyos-native proton-cachyos-slr

MangoHud

# ~/.config/MangoHud/MangoHud.conf
hud_compact
hud_no_margin
text_outline
background_alpha=0.3
font_size=14
table_columns=2
fps
frametime
gpu_stats
gpu_temp
gpu_core_clock
gpu_mem_clock
vram

cpu_stats
cpu_temp
cpu_mhz

wine

No frametime graph. It's a massive visual for one number. Compact two-column layout shows everything at a glance.

OptiScaler

DLSS-style upscaling on non-RTX hardware via FSR injection. Install as Wine DLL override per-game. OptiScaler on GitHub.

OnlineFix / cracked games

# Wine DLL overrides
onlinefix64=n;winmm=n,b;steam_api64=n;version=n,b

contents

11. Audio: PipeWire low latency

# /etc/pipewire/pipewire.conf.d/10-low-latency.conf
context.properties = {
    default.clock.rate            = 48000
    default.clock.quantum         = 128
    default.clock.min-quantum     = 128
    default.clock.max-quantum     = 512
}

Latency: 128 / 48000 ≈ 2.67ms per cycle. Round-trip ≈ 5.3ms + hardware.

Tradeoff: Lower quantum = lower latency, higher CPU. 128 is good for competitive/rhythm games. 256 (~5.3ms) is fine for general desktop.

systemctl --user restart pipewire
pw-top  # check QUANT column during playback

CachyOS already provides: rtkit daemon, audio group rtprio 99, CPU DMA latency device access, snd_hda_intel power save disabled on AC, HPET/RTC permissions for audio group.

contents

12. Network

# In /etc/sysctl.d/99-latency.conf:
net.core.netdev_max_backlog = 16384
net.ipv4.tcp_fastopen = 3
net.core.somaxconn = 1024

CachyOS already provides: fq_codel qdisc, cubic TCP congestion, 120s keepalive, Cloudflare/Google NTP, systemd-resolved DNS.

contents

13. System limits & sysctl

File descriptors

Wine/Proton opens many FDs. CachyOS defaults are generous but can be raised further.

# /etc/security/limits.conf
* hard core 0
* soft core 0
* hard nofile 524288
* soft nofile 524288
root hard nofile 524288
root soft nofile 524288

Complete sysctl

# /etc/sysctl.d/99-latency.conf
vm.swappiness = 180
vm.dirty_background_ratio = 3
vm.dirty_ratio = 8
net.core.somaxconn = 1024
net.core.netdev_max_backlog = 16384
net.ipv4.tcp_fastopen = 3
fs.file-max = 2147483647
fs.inotify.max_user_watches = 524288
fs.inotify.max_user_instances = 1024

sudo sysctl --system

contents

14. Services to disable

These are enabled by default on CachyOS. They're useless on a dedicated gaming desktop.

contents

15. Desktop & Wayland

A Wayland compositor with fullscreen bypass (direct scanout) hands the display buffer directly to the GPU when a game goes fullscreen, with no compositing step and no added latency.

Good options: COSMIC, KWin (KDE Plasma), Mutter (GNOME). All support this.

NVIDIA + Wayland requires: nvidia_drm.modeset=1 in boot parameters. Without it, Wayland sessions fall back to software rendering or fail.

Reality check: If you have driver < 550.xx or a compositor without explicit sync, Wayland will be a flickery mess. Use X11.

Gamescope: Skip it unless you need FSR upscaling, HDR tonemapping, or nested sessions. A native compositor with fullscreen bypass is faster.

contents

16. Minimum viable tuning (The 80/20)

Don't want to apply everything? These 4 things give 90% of the gain with 10% of the effort.

Do these and you'll solve 80% of the stutter and latency issues. The rest is fine-tuning.

contents

17. Verification checklist

# ── Boot ──
cat /proc/cmdline
# ── CPU ──
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor    # → performance
cat /sys/devices/system/cpu/cpu0/cpufreq/energy_performance_preference  # → performance

# ── Kernel ──

zcat /proc/config.gz | grep CONFIG_PREEMPT    # → CONFIG_PREEMPT=y
zcat /proc/config.gz | grep CONFIG_HZ=        # → CONFIG_HZ=1000
zcat /proc/config.gz | grep CONFIG_SCHED_CLASS_EXT   # → =y
uname -r                                      # "-bore" in the string = BORE kernel, otherwise EEVDF
scxctl get                                    # → currently loaded SCX scheduler, or "none" for EEVDF/BORE

# ── Memory ──

zramctl                          # check compression, size
sysctl vm.swappiness             # → 150
sysctl vm.dirty_background_ratio vm.dirty_ratio  # → 3 and 8

# ── Storage ──

cat /sys/block/sdX/queue/scheduler  # → [adios] for SSD, [bfq] for HDD

# ── GPU ──

nvidia-smi -q | grep Persistence         # → Enabled
nvidia-smi -q -d CLOCK | grep "Max Clocks"

# ── Audio ──

pw-top   # QUANT = 128 during playback

# ── Network ──

sysctl net.ipv4.tcp_fastopen           # → 3
sysctl net.core.netdev_max_backlog     # → 16384

# ── Limits ──

ulimit -n   # → 524288

# ── Gamemode (while game is running) ──

gamemoded -s   # → "gamemode is active"

# ── Services ──

systemctl is-active ananicy-cpp power-profiles-daemon  # → inactive

contents

18. System recovery (if you break it)

If your system doesn't boot after changing kernel parameters:

1. Boot from USB (CachyOS installer ISO).
2. Mount your root partition: mount /dev/sdXY /mnt (and /boot if separate).
3. Chroot in: arch-chroot /mnt.
4. Revert your changes: restore /etc/default/limine.bak or remove the offending parameter, then limine-mkinitcpio.
5. Reboot.

Alternatively, if you used BTRFS snapshots (@pre-optimization), roll back from the bootloader menu or rename subvolumes from the chroot.

contents

19. Configuration file reference

All paths absolute from / unless noted. Create these files, apply, reboot.

/etc/default/limine

KERNEL_CMDLINE[default]+="quiet mitigations=off nowatchdog nmi_watchdog=0 nvidia_drm.modeset=1 tsc=reliable clocksource=tsc intel_pstate=active preempt=full split_lock_detect=off pcie_aspm=performance intel_idle.max_cstate=1 transparent_hugepage=madvise splash rw root=UUID=..."

Apply: sudo limine-mkinitcpio && sudo reboot

/etc/scx_loader/config.toml

default_sched = "scx_bpfland"
default_mode = "Gaming"

Apply: sudo systemctl enable --now scx_loader.service

/etc/tmpfiles.d/force-performance.conf

w /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor - - - - performance

/etc/tmpfiles.d/force-epp-performance.conf

w /sys/devices/system/cpu/cpu*/cpufreq/energy_performance_preference - - - - performance

/etc/sysctl.d/99-latency.conf

vm.swappiness = 180
vm.dirty_background_ratio = 3
vm.dirty_ratio = 8
net.core.somaxconn = 1024
net.core.netdev_max_backlog = 16384
net.ipv4.tcp_fastopen = 3
fs.file-max = 2147483647
fs.inotify.max_user_watches = 524288
fs.inotify.max_user_instances = 1024

Apply: sudo sysctl --system

/etc/security/limits.conf

* hard core 0
* soft core 0
* hard nofile 524288
* soft nofile 524288
root hard nofile 524288
root soft nofile 524288

/etc/security/limits.d/10-gamemode.conf

@gamemode - nice -10

/etc/environment

__GL_THREADED_OPTIMIZATION=1
__GL_SHADER_DISK_CACHE=1
__GL_SHADER_DISK_CACHE_SKIP_CLEANUP=1
__GL_VRR_ALLOWED=0
__GL_SYNC_TO_VBLANK=0
__GL_MaxFramesAllowed=1
VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/nvidia_icd.json
WINEDEBUG=-all

/etc/systemd/zram-generator.conf.d/override.conf

[zram0]
compression-algorithm = lz4

Apply: sudo systemctl restart systemd-zram-setup@zram0

/etc/udev/rules.d/60-ioschedulers.rules

# HDD: BFQ
ACTION=="add|change", KERNEL=="sd[a-z]*", ATTR{queue/rotational}=="1", \
    ATTR{queue/scheduler}="bfq"
# SSD: ADIOS
ACTION=="add|change", KERNEL=="sd[a-z]*|mmcblk[0-9]*", ATTR{queue/rotational}=="0",
ATTR{queue/scheduler}="adios"

# NVMe: ADIOS

ACTION=="add|change", KERNEL=="nvme[0-9]*", ATTR{queue/rotational}=="0",
ATTR{queue/scheduler}="adios"

Apply: sudo udevadm control --reload-rules && sudo udevadm trigger

/etc/pipewire/pipewire.conf.d/10-low-latency.conf

context.properties = {
    default.clock.rate            = 48000
    default.clock.quantum         = 128
    default.clock.min-quantum     = 128
    default.clock.max-quantum     = 512
}

Apply: systemctl --user restart pipewire

/etc/gamemode.ini

[general]
renice = 20
ioprio = 0
[gpu]
apply = 1
nv_powermizer_mode = 1
[custom]
start = nvidia-smi -pm 1
end = nvidia-smi -pm 0

/etc/X11/xorg.conf.d/20-nvidia-full-rgb.conf

Section "Device"
    Identifier "NVIDIA"
    Driver "nvidia"
    Option "ColorSpace" "RGB"
    Option "ColorRange" "Full"
EndSection

/etc/fstab (example)

UUID=<boot>  /boot  vfat  defaults,umask=0077  0 2
UUID=<root>  /      xfs   defaults,lazytime,noatime,inode64,logbsize=256k,noquota  0 1
tmpfs        /tmp   tmpfs defaults,noatime,mode=1777  0 0
UUID=<hdd>   /mnt/data  xfs  defaults,noatime,nofail,allocsize=64m  0 0
UUID=<ssd>   /mnt/scratch  f2fs  defaults,noatime,nofail  0 0

~/.config/MangoHud/MangoHud.conf

hud_compact
hud_no_margin
text_outline
background_alpha=0.3
font_size=14
table_columns=2
fps
frametime
gpu_stats
gpu_temp
gpu_core_clock
gpu_mem_clock
vram

cpu_stats
cpu_temp
cpu_mhz

wine

contents