Release: 0.8.5
Recommended nVidia proprietary driver branch 550+ for best compatibility with OptiX and CUDA.
CUDA SDK up to 12.9 should now mostly be supported. There can still be missing functions
Up to OptiX ABI 105 (SDK 9.0.0) should be supported. Additional logging: WINEDEBUG=+nvoptix
Older drivers that does not support CUDA SDK 10 and older may also fail/crash. Report any crashes with
logs on the issue page. However these are highly outdated drivers in general and will probably not
see too much attention in that regard.
Create logs typically with: WINEDEBUG=-all,+nvcuda,+nvml wine ./yourapp.exe > yourapp.log 2>&1
If you suspect errors with video encoding/decoding (NVENC), you can create a log like this:
WINEDEBUG=-all,+nvcuvid,+nvencodeapi wine ./yourapp.exe > yourapp.log 2>&1
Also see notes about DXVK-NVAPI below for additional logging.
Library contains:
nvapi (dxvk-nvapi + additions - https://github.com/SveSop/dxvk-nvapi/tree/experimental)
nvcuda (wine-staging + development - https://github.com/SveSop/nvcuda)
nvcuvid/nvencodeapi (wine-staging + development - https://github.com/SveSop/nvenc)
nvml (wine-nvml - https://github.com/Saancreed/wine-nvml)
nvoptix (wine-nvoptix - https://github.com/SveSop/wine-nvoptix)
- [WINE] (version >= 10.0) https://www.winehq.org/
- [Meson] (version >= 0.61) http://mesonbuild.com/
- [NINJA] https://ninja-build.org/
- [MINGW] (version >= 10.0) https://www.mingw-w64.org/
- [GCC/G++] (version >= 12.0) https://gcc.gnu.org/
./package-release.sh name destdir [--no32]
The --no32 build option only build 64-bit version of nvcuda/nvenc/nvml, but still
maintain the 32-bit version of nvapi.dll.
You can install symlinks and dll-overrides to the libraries in your wineprefix like this:
WINEPREFIX=/your/wine/prefix ./setup_nvlibs.sh install
To uninstall:
WINEPREFIX=/your/wine/prefix ./setup_nvlibs.sh uninstall
Check out README file usage of DXVK-NVAPI.
https://github.com/SveSop/dxvk-nvapi/blob/master/README.md
Typical log level usage:
DXVK_NVAPI_LOG_LEVEL=none (Default) No output.
DXVK_NVAPI_LOG_LEVEL=info Informational output of function calls.
DXVK_NVAPI_LOG_LEVEL=trace Detailed debug information.
For misbehaving games that do not work too well with DLSS or otherwise do not work with
the detected architecture, you can override detected architecture like this:
DXVK_NVAPI_GPU_ARCH= with one of the following options:
GK100 (Kepler)
GM200 (Maxwell)
GP100 (Pascal)
GV100 (Volta)
TU100 (Turing)
GA100 (Ampere)
AD100 (Ada)
GB200 (Blackwell)
Known games that may need override:
Monster Hunter World : DXVK_NVAPI_GPU_ARCH=GP100
War Thunder : DXVK_NVAPI_GPU_ARCH=GP100
Ghost of Tsushima : DXVK_NVAPI_GPU_ARCH=GA100
This includes the Vulkan Reflex layer in /layer subfolder.
Read about usage here:
https://github.com/jp7677/dxvk-nvapi#vulkan-reflex-layer
Various DLSS settings has been added to dxvk-nvapi. You can read more about various settings here:
https://github.com/jp7677/dxvk-nvapi/wiki/Passing-driver-settings
In order to use DXVK-NVAPI (NvAPI) with some extra GPU information (temps and the likes)
you need to copy the NVML binaries to your wine binaries folder. You can do this by
running the included nvml_setup.sh script.
See the Readme_nvml.txt file.
You can run tests of your adapter and functions by running the test case file:
WINEPREFIX=/your/wine/prefix wine bin/nvapi64-tests.exe
and you can run some small CUDA tests to verify your CUDA adapter like this:
WINEPREFIX=/your/wine/prefix wine bin/cudatest.exe
PS. You must install said libraryfile in your WINEPREFIX (see above)
Added a script that will install the libraries in a bottle.
This assumes you are using bottles installed via FlatPak.
Run ./bottles-install.sh for a list of your available bottles, and use it like this:
./bottles-install.sh YourBottle to install the libraries in the YourBottle bottle.
You will want to run the script ever time you upgrade nvidia-libs binaries, and possibly
if you change/update your runner. Eg. switching from Caffe-9.7 to Caffe-9.8.
The 32-bit versions of nvcuda.dll, nvencodeapi.dll, nvcuvid.dll and nvml.dll all requires
that the 32-bit driver libraries are installed. Depending on distro these can have separate
packages that needs to be installed with the driver.
You can drop-in the libraries in Proton8 - Experimental or GE-Proton-8 or newer like this:
PROTON_LIBS='$HOME/.steam/compatibilitytools.d/Proton-8.27-GE' ./proton_setup.sh
This will replace or add the libs to Proton for use with games.
The installscript is no longer compatible with older Proton-6.3 or Proton-GE-6/7 versions.
OBS!
You can add:
"DXVK_ENABLE_NVAPI": "1",
"PROTON_ENABLE_NVAPI": "1",
to your user_settings.py script in the proton folder eg:
PROTON_LIBS='$HOME/.steam/compatibilitytools.d/GE-Proton-8.27/user_settings.py'
Or you can run the game with PROTON_ENABLE_NVAPI=1 DXVK_ENABLE_NVAPI=1 %command%
Some additional tweaks may be required depending on game used and version of Proton.
It is HIGHLY RECOMMENDED that you re-create the game-prefix after installing the nvidia-libs
package, to make sure they are used in the new prefix. The game prefixes are typically located
in $HOME/.steam/steam/steamapps/compatdata/XXXXXXX where XXXXXXX is the steam AppID.
OpticalFlow for use with DLSS seems to have been discontinued with NVIDIA DLSS4.
If OpticalFlow is required for some app or program, there is a archived version that can be
used in the nvofapi-workaround branch if needed. This will however build an old version
of nvapi & friends, and will (for now) not be updated.
In case you do NOT use nvml (wine-nvml) library to emulate nvml.dll, DAZ Studio MAY
show 2 "adapters" for rendering. Choose the one that show your GPU name, and NOT the other.
NVML can be installed by using the setup_nvml.sh script included in the package and for
DAZ Studio it is recommended.
Some older SDK samples may require specifying LD_LIBRARY_PATH to the system folder
where libcuda.so.1 and other libnvidia-xxx.so libraries reside. This varies depending
on distro. Ubuntu typicall uses /lib/x86-64-linux-gnu/ for 64-bit and
/lib/i386-linux-gnu/ for 32-bit.
For instance to get the NVIDIA SuperSonic Sled demo to run on Ubuntu you can load wine
with using LD_LIBRARY_PATH=/lib/i386-linux-gnu:$LD_LIBRARY_PATH
Loads of reference info here:
Tool to check various GPU options:
http://www.ozone3d.net/gpu_caps_viewer/