Nvfancontrol provides dynamic fan control for NVidia graphic cards on Linux and Windows.
Sometimes it is desirable to control the fan speed of the graphics card using a custom response curve instead of the automatic setting that is built into the card's BIOS. Especially in newer GPUs the fan does not kick in below 60°C or a certain level of GPU utilization. This is a small toy project in Rust to achieve a more elaborate control over this using either XNVCtrl in Linux or NVAPI in Windows. It is a work in progress so proceed with caution!
The minimum supported driver version is 352.09. For GPUs with multiple independent cooler control nvfancontrol will autodetect and apply the provided response curve to each of the available fans separately.
Pre-built binaries for the latest release are provided however if you want to build the project from source read along.
Prerequisites for Linux
You will need:
- the Rust compiler toolchain, stable >=1.34 or nightly (build)
- XNVCtrl; static (build only) or dynamic (build and runtime)
- Xlib (build and runtime)
- Xext (build and runtime)
Since XNVCtrl supports FreeBSD in addition to Linux these instructions should also work for FreeBSD without further modifications. However nvfancontrol is completely untested on FreeBSD (bug reports are welcome).
Prerequisites for Windows
You will need:
- the Rust compiler toolchain, stable >=1.15 or nightly. Be adviced that you need the MSVC ABI version of the toolchain not GNU. In order to target the MSVC ABI for Rust you will also need the Visual C++ build tools package or any recent version of Visual Studio (2015+). If you are using rustup (which you should) you will be warned about this (build only)
- the NVAPI libraries (build only).
Depending on which version you are building (x86, x64 or both) place
nvapi64.libor both in the root of the repository. As
nvapiis linked statically there are no runtime dependencies apart from the NVidia driver.
For both platforms run
cargo build --release. Upon successful compilation the
executable can be found in
target/release/nvfancontrol. On Linux the build
tool expects the libraries installed in
/usr/local/lib. In case
you have libraries installed in different locations export them using the
LIBRARY_PATH environment variable (colon separated paths). By default
libXNVCtrl will be linked statically. If a static version of
not available or you explicitly want it to be linked dynamically add
--features=dynamic-xnvctrl to the
Enable Coolbits (Linux only)
For Linux ensure that Coolbits is enabled from your X11 server settings. To
do so create a file named
/usr/share/X11/xorg.conf.d/ (depends on distribution) containing the
Section "Device" Identifier "Device 0" Driver "nvidia" VendorName "NVIDIA Corporation" BoardName "IDENTIFIER FOR YOUR GPU" Option "Coolbits" "4" EndSection
The important bit is the
Coolbits option. Valid Coolbits values for dynamic
fan control are
12. A sample configuration file is provided.
Use and configure
To run the program just execute the
nvfancontrol binary. Add the
--debug argument for more output. To add a custom curve you can provide a
custom configuration file. On Linux create a file named
under the XDG configuration directory (
/etc/xdg for per-user
and system-wide basis respectively). On Windows create the file in
C:\Users\[USERNAME]\AppData\Roaming instead. The configuration file should
contain pairs of whitespace delimited parameters (Temperature degrees Celsius,
Fan Speed %).
30 20 40 30 50 40 60 50 70 60 80 80
Lines starting with
# are ignored. You need at least two pairs of values.
Bear in mind that for most GPUs the fan speed can't be below 20% or above 80%
when in manual control, even if you use greater values. However, since these
limits are arbitrary and vary among different VGA BIOS you can override it
--limits option. For example to change the limits to 10%
and 90% pass
-l 10,90. To disable the limits effectively enabling the whole
range just pass
-l 0. In addition note that the program by default will not
use the custom curve if the fan is already spinning in automatic control. This
is the most conservative configuration for GPUs that turn their fans off below
a certain temperature threshold. If you want to always use the custom curve
pass the additional
--force argument. To terminate nvfancontrol send
a SIGINT or SIGTERM on Linux or hit Ctrl-C in the console window on Windows.
Although presently nvfancontrol is limited to a single GPU, users can select
the card to modulate the fan operation using the
--gpu switch. GPUs
are indexed from
0. To help with that option
list all available GPUs with their respective coolers. On Windows coolers are
0 for each GPU. On Linux each available cooler on the system is
assigned a unique id.
Third party interfacing
nvfancontrol offers two ways to dump the output of the program for integration
with third party programs. Using the
-j option a JSON represantation of the
current data is printed to
stdout. As all other messages are printed to
stderr the data can be parsed by reading new-line delimited data from the
stdout. If this is not desirable a builtin TCP server is also
provided which can be enabled using the
-t option. This option can optionally
be followed by a port number (default port is 12125). The server prints the
JSON data through the socket and immediately closes the connection. The message
is always terminated with a new-line character.
Bugs and known issues
Although nvfancontrol should work with most Fermi or newer NVidia cards it has
been tested with only a handful of GPUs. So it is quite possible that bugs or
unexpected behaviour might surface. In that case please open an issue in the
bug tracker including the complete program output (use the
RPM reporting for GPUs with multiple fans on Windows is incorrect or totally
wrong because the provided function
NvAPI_GPU_GetTachReading is limited to a
single fan. There is nothing in the public NVAPI to suggest otherwise.
However, speed in % should work as expected. In any case multiple cooler
support on Windows is not thoroughly tested so bug reports are always welcome!
As mentioned before, nvfancontrol is limited to a single (but selectable) GPU. The underlying code does support multiple GPUs but exposing this support to the user-facing program will require possibly breaking alterations to the configuration file. It will be added eventually.
This project is licensed under the GPLv3 or any newer.