Skip to content

Instantly share code, notes, and snippets.

/*
CUDA perfboost bypass, distributed under MIT License
-- @Ristovski
Each CUDA session is boosted to a fixed clock even when no real compute is being done,
this in turn causes the GPU to run hotter, sometimes at considerably higher power.
This preloadable library bypasses that by nuking the NV2080_CTRL_PERF_BOOST ioctl.
Mainly useful for mpv/other video players that utilize hardware video decode (nvdec),
as unlike on Windows, nvdec on linux utilizes CUDA for nvdec<->OpenGL/Vulkan interop.
#
# Automatically generated file; DO NOT EDIT.
# Linux/arm64 5.4.242 Kernel Configuration
#
CONFIG_CC_VERSION_TEXT="clang version 16.0.6"
CONFIG_GCC_VERSION=0
CONFIG_CC_IS_CLANG=y
CONFIG_LD_IS_LLD=y
CONFIG_CLANG_VERSION=160006
CONFIG_CC_CAN_LINK=y

Keybase proof

I hereby claim:

  • I am ristovski on github.
  • I am ristovski (https://keybase.io/ristovski) on keybase.
  • I have a public key whose fingerprint is 7324 80DB 52DE 1DB4 AE3A 6ECB 9857 35F1 EF68 DDEC

To claim this, I am signing this object: