[GUIDE] Running CUDA 12.9 on NVIDIA Pascal (GTX 10xx) with Arch Linux
Tested on: X570DD_M570DD / GTX 1050 mobile / Arch Linux x86_64 / NVIDIA 580.178.04 driver / CUDA V12.9.86
What happened: As everyone knows, NVIDIA dropped the Pascal line in the 590.xxx.xx+ and CUDA 13.x+ driver line, and if you try to install the nvidia-open (now nvidia) or cuda package, you will encounter the fact that Pascal has been cut from the new versions.
Installing the NVIDIA 580xx Driver: To ensure you have the latest driver that supports Pascal, we'll need the AUR Helper to download the following packages via AUR. I'll be using yay:
yay -S lib32-nvidia-580xx-utils
yay -S nvidia-580xx-dkms
yay -S nvidia-580xx-settings
yay -S nvidia-prime
yay -S opencl-nvidia-580xx
Installing the GCC14: since CUDA 12.9 requires a compatible GCC version, we install GCC 14 for the toolkit. Since compiling GCC 14 manually takes a long time, we’ll simply download it from the Arch archives:
cd /tmp
wget https://archive.archlinux.org/packages/g/gcc14/gcc14-14.3.1+r416+g44d5743651c4-2-x86_64.pkg.tar.zst
wget https://archive.archlinux.org/packages/g/gcc14/gcc14-libs-14.3.1+r416+g44d5743651c4-2-x86_64.pkg.tar.zst
sudo pacman -U gcc14-*.pkg.tar.zst
Next, to verify that gcc14 has indeed been installed, you can run the following command:
gcc-14 --version
g++-14 --version
which gcc-14
Next, we need to exclude gcc14 and gcc14-libs from being updated by our AUR helper. Change the following lines in /etc/pacman.conf:
IgnorePkg = gcc14 gcc14-libs
Simply add gcc14 and gcc14-libs to your IgnorePkg
Installing the CUDA 12.9:
There's a great package called cuda-pascal, and we'll install it because it's designed specifically for our purpose:
yay -S cuda-pascal
Next, add CUDA to your PATH. This is a temporary command, you'll need to add CUDA to your shell's PATH!
export PATH=/opt/cuda/bin:$PATH
Now you can run a test on the NVCC compiler that we just installed:
nvcc --version
echo '#include <cstdio>
__global__ void hello() {
printf("Hello from GPU!\n");
}
int main() {
hello<<<1, 1>>>();
cudaDeviceSynchronize();
}' > /tmp/hello.cu && nvcc /tmp/hello.cu -o /tmp/hello && /tmp/hello
If you see the message “Hello from GPU!”, congratulations! You've successfully installed CUDA!
Software configuration:
- Blender:
- To have Blender use our configured CUDA, we need to open Blender, press ‘Ctrl+,’, go to the ‘System’ section, select ‘CUDA’ in the Cycles rendering section, and choose GPU
Then restart Blender if necessary.
Now go to the ‘Render’ section in the sidebar and select Cycles and GPU Compute.
Set Viewport Shading to Rendered.
When you launch Blender for the first time, you should see “Loading render kernels” appear in the top-left corner.
Wait a moment, and your image should render using your Pascal GPU via CUDA in Cycles!
You can verify this using nvidia-smi or ps aux | grep blender
-
Ollama:
The Ollama packages from the Arch repositories and AUR did not work correctly on my GTX 1050, so I used the official installation script instead:
curl -fsSL https://ollama.com/install.sh | sh
Once you've installed ollama, you can run the phi4-mini test model (lightweight and smart):
ollama run phi4-mini --verbose
In another terminal, run `ollama ps`
and you should see the line “100% GPU” in the “PROCESSOR” column. If so, congratulations - you've successfully run ollama on your Pascal graphics card!
if you want to test the newest UI try out v1.2.0-beta.1 here: https://github.com/adityakrishnan005-a11y/ScreenGuard/releases/tag/v1.2.0-beta.1
