Hello, I would really love to use my new 5060 on Arch, Thank you.
You can do, install the latest driver from nvidia download page. Anyway, if You are on arch install using:
sudo pacman -S nvidia-open-dkms
Yeah, but you don’t need that because Arch repos include nvidia drivers.
This is incorrect. There’s no driver for 5060 available for linux at the moment. The latest card that is supported on linux is the 5060 TI through driver version 575.51.02 (nvidia-open-beta on arch repo). You can check official kernel module repository on github for list of supported hardware.
575.51.02 WILL see the card but will not correctly identify its model. The card only offers very rudimentary functionality on linux at this point.
It took like 2 weeks for a 5070 driver, so expect the same or more for the 5060.
The 5060 is supported now with Driver Details | NVIDIA 575.57.08
hi there, I bought an MSI gaming trio GeForce rtx 5060 TI OC edition. compared to my old Nvidia GTX 1660 super Asus tuf, the card is super quiet. However (I am under Gentoo) if I install the
nvidia-drivers-575.57.08
then I get the following log on the X server
Nvidia glx module 575.57.08
The Nvidia GPU at PCI:45:0:9 is not supported by 575.57.08
failed to initialize Nvidia graphics device
unloading module Nvidia
You need to have USE flag kernel-open – Gentoo Packages set for Blackwell.
under windows the card is recognized and works of course. But I want to use it for cuda unter Linux…
so it would be really nice if that new driver would recognize my gpu
hi thanks. I will try it out…
worked like a charm. Many thanks.
now I have to change the variables in /etc/portage/make.conf
CUDAARCHS=12
TORCH_CUDA_ARCH_LIST=12.0
and recompile some software in order to take the new cuda capabilities into account.
How do I set up GCC such that it uses the newest Ptx and cudaarch for openacc and openmp? Currently I use some default, which is a bit old.
I guess also I have to reorganize he hard drives of my pc a bit so that my old Nvidia 1660 super can be installed together with the 5060 ti. I would like to see how I can write c++ programs that use arrays of several GPUs. I guess that openmpi processes that start open acc or openmp functions would be optimal for that, because then the GPUs can be on several different computers…
for any mathematician, having two large matrices is always better than having just one matrix…
I am not (anymore) familiar with Gentoo. However I remember they have an excellent documentation and forums, maybe you should ask there when it comes to distro-specific settings. I think there are doc pages on Gentoo site that detail compiler options/config. I am happy to check though.
Did you emerge dev-util/nvidia-cuda-toolkit – Gentoo Packages already?
Hi Gentoo has excellent documentation indeed. But only for genrok specific things.
How to set up the GCC compiler such that it uses the most recent cuda capability it supports with its nvptx backend is not anymore specific to Gentoo, but to the gnu compiler site. I found nothing on this however.
Also, how does this work with cuda by the way, if one has two cards and one card has capability 7.5 and the other cuda capability 12 and both work with the same Nvidia driver on the same pc?
Must I then set GCC to compile my openacc program for the smallest cuda version? Probably? How does this work with clang? or NVC++ and nvcc?
By the way since my software needed some c++ and openacc/openmp support, which turned out to lead to frequent internal compiler errors, I filed several bugs for GCC 14.
With version 15.1 most of them are fixed now.
GCC 15.1 should work much better with openacc and openmp and C++ than before
I will now continue to develop my software and see if I find further bugs…