From 8849eedd06107aca5fe4602e1c5c0bf0c97a3d8d Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Sveinar=20S=C3=B8pler?= Date: Fri, 24 May 2024 20:18:46 +0200 Subject: [PATCH] Release v0.7.17 Add support for CUDA SDK 12.5 Add helper library for FFT (cufft64_11.dll) when running FFT functions from CUDA SDK 12.3 or newer. Add 32-bit helper library for cuda RunTime 3.0 (cudart32_30_14.dll) for running old CUDA SDK 3.0 samples and NVIDIA demo's. This however requires some handheld fiddling. See README.md for info. Various bugfixes and tweaks. --- README.md | 29 +++++++++++++++++++++++++++-- 1 file changed, 27 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index af2f044..1b539ed 100644 --- a/README.md +++ b/README.md @@ -1,10 +1,11 @@ # NVIDIA Libs -Release: 0.7.16 +Release: 0.7.17 Recommended nVidia proprietary driver branch 550+ for best compatibility with OptiX and CUDA. CUDA SDK >11.5 should now mostly be supported. There can still be missing functions -Running CUFFT samples with CUDA 12.3 or higher is known to crash. +Added CUFFT helper library for FFT functions when running SDK 12.3 and newer +See https://github.com/SveSop/nvidia-libs#helper-lib-usage below for usage Older drivers that does not support 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 However these are highly outdated drivers in general and will probably not @@ -68,6 +69,30 @@ Or you can run the game with `PROTON_ENABLE_NVAPI=1 DXVK_ENABLE_NVAPI=1 %command Some additional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. +## Helper lib usage +To use FFT helper library `cufft64_11.dll` when running CUDA 12.3 or newer, simply +symlink the binary to the same folder where the executable you are running like this: +`ln -sf nvidia-libs-release\lib64\wine\x86_64-unix\cufft64_11.dll.so cufft64_11.dll` +NB. You should probably make a backup of the original `cufft64_11.dll` in that folder. +This however requires CUDA SDK 12.3 or newer to be installed on your system! + +To use CUDART 3.0 helper library `cudart32_30_14.dll` when running old SDK 3.0 samples +or demos like NVIDIA SuperSonic Sled Demo, do the same as above: +`ln -sf nvidia-libs-release\lib\wine\i386-unix\cudart32_30_14.dll.so cudart32_30_14.dll` +This is ONLY for 32-bit and requires the 32-bit version of the CUDA SDK 3.0 toolkit +installed/extracted on your system (see below). + +Newest CUDA SDK toolkit can be downloaded here: +https://developer.nvidia.com/cuda-downloads + +The CUDA SDK 3.0 toolkit can be downloaded here: +https://developer.download.nvidia.com/compute/cuda/3_0/toolkit/cudatoolkit_3.0_linux_32_ubuntu9.04.run + +PS. Yes, it is quite a bit quirky to get SDK 3.0 installed like this, but you should be able to +extract the lib* files, and you can load them with `LD_LIBRARY_PATH` to get it working on newer +distros. Due to this somewhat troublesome exercise, these helper binaries are built but not +installed into the WINEPREFIX with the installscript as this is mostly for "specially interested". + ## Info Loads of reference info here: