On Manjaro’s stable branch, cuda version is 12.6.3. However, the driver version is 550.144.03-1 which only support CUDA toolkit 12.4 according to Nvidia’s documentation. Running nvidia-smi shows tha CUDA version is 12.4, and you run nvcc --version and it shows that the CUDA toolkit version is 12.6.3.
I’m not sure if this cause the following issue, but when I build and run mistral.rs I got the following error
Error: DriverError(CUDA_ERROR_UNSUPPORTED_PTX_VERSION, "the provided PTX was compiled with an unsupported toolchain.")
which caused by the mismatch between the versions of CUDA toolkit and driver.
I solve this issue by downgrading the cuda and cuda-tools to 12.4.1-4.
With Nvidia you are likely better of using Manjaro Edge (unstable) branch as this (hopefully) should match better.
Hi manjaro team. I also met this issue these days, and now see this topic.
Would you (manjaro team) tell us (manjaro users) whether manjaro have plans about switching to the newest drivers on stable branch?
As the latest stable snap did not include Nvidia 570, I assume it was not deemed stable by @philm, who acts as the Manjaro release manager.
Last I checked the Nvidia 570 series was a beta driver
Only the release manager can answer that question
Current status 2025-03-19T06:55:00Z
$ mbn info linux612-nvidia -q | grep -e Version -e Branch
Branch : unstable
Version : 570.124.04-4
Branch : testing
Version : 550.144.03-9
Branch : stable
Version : 550.144.03-9
If you need the latest version for for specific tasks the previous recommendation still stands stands.
Switch to unstable branch
sudo pacman-mirrors --api --set-branch unstable
sudo pacman -Syu
Or wait
→ https://www.phoronix.com/news/NVIDIA-570.133.07-Linux-Driver
I don’t actually need the newest driver. I just feel unsmooth if I have to use the “unstable” drivers in order to get my cuda programs compiled and run them well.
In other word, I hope that manjaro team can get the cuda version and driver version matched (older cuda or newer driver?), so that we users don’t have to suffer from the issue.
To mitigate the issue locally, I downgraded my cuda version just now 
As of 2025-03-20T11:18:00Z
$ mbn info nvidia-dkms -q nvidia-dkms | grep -e Version -e Branch
Branch : archlinux
Version : 570.124.04-1
Branch : unstable
Version : 570.133.07-2
Branch : testing
Version : 570.133.07-2
Branch : stable
Version : 550.144.03-1
→ [Testing Update] 2025-03-20 - Kernels, Mesa 25.0.1, Frameworks 6.12, Pipewire 1.4.1
News: manjaro has bumped nvidia driver to 570 just now on stable branch, we can now use cuda 12.8.