Skip to content

Kernel & Performance

The system runs the CachyOS kernel with low-latency tuning, set in nixos/configuration.nix via the nix-cachyos-kernel overlay:

boot.kernelPackages = pkgs.cachyosKernels.linuxPackages-cachyos-latest;

Kernel parameters

Parameter Purpose
preempt=full Full preemption for interactivity under load
threadirqs Threaded IRQ handling
processor.max_cstate=1 Limit CPU C-states (Ryzen hybrid-GPU stability)
nvidia-drm.modeset=1 NVIDIA DRM modesetting
nvidia.NVreg_EnableGpuFirmware=1 Enable GSP firmware
amdgpu.dcdebugmask=0x10 AMD display-core stability workaround
amdgpu.gpu_recovery=1 Enable AMD GPU recovery
sysrq_always_enabled=1 Magic SysRq available

Sysctl tuning

boot.kernel.sysctl sets memory, scheduling, and network parameters:

"vm.max_map_count" = 2147483642;   # very high mmap limit (large apps / games)
"vm.swappiness" = 180;             # aggressive swap (paired with zram-style setups)
"vm.page-cluster" = 0;
"vm.vfs_cache_pressure" = 50;

"net.core.default_qdisc" = "fq";
"net.ipv4.tcp_congestion_control" = "bbr";
"net.ipv4.tcp_fastopen" = 3;
"net.ipv4.tcp_slow_start_after_idle" = 0;
"net.core.netdev_max_backlog" = 16384;
"net.core.somaxconn" = 8192;

"kernel.panic" = 10;               # reboot 10s after panic
"kernel.panic_on_oops" = 1;
"kernel.sched_cfs_bandwidth_slice_us" = 3000;

Service-manager timeouts

To keep shutdown fast (a recurring pain point with FUSE mounts and stubborn units), systemd.settings.Manager sets DefaultTimeoutStopSec = "10s" (5s for the user manager), and a decapitate-fuse-mounts oneshot force-unmounts the xdg-document-portal FUSE at shutdown to release /nix.

Schedulers

services.scx (sched-ext, e.g. scx_bpfland) is wired in but currently disabled (enable = false); the CachyOS kernel's default scheduler is in use.