I’m starting to study parallel programming and would like to use the nvidia drivers. In my case, I have a GeForce 820M video card with drivers 390.157, the latest for this hardware. I also tried to install the CUDA version for this hardware and driver configuration that I have. In this case, from what I could see at https://forum.manjaro.org/t/cuda-jl-only-supports-nvidia-drivers-for-cuda-10-1/36573/5 it seems that the latest version would be CUDA 9.1. I installed version 9.0, version 9.1 was not available in the AUR (I use Arch Linux). However, when running the example from /opt/cuda-9.0/samples/0_Simple/vectorAdd I got:
Copy input data from the host memory to the CUDA device
CUDA kernel launch with 196 blocks of 256 threads
Failed to launch vectorAdd kernel (error code no kernel image is available for execution on the device)!
When searching, I found this post https://stackoverflow.com/questions/72270546/cuda-failed-to-launch-kernel-no-kernel-image-available-for-execution In other words, it seems that support for the Fermi class has been removed, and the latest version available would be CUDA 8.0.
So, I proceeded with the installation (yay -S cuda-8.0) this time, when running /opt/cuda/samples/0_Simple/vectorAdd/./vectorAdd I got:
Copy input data from the host memory to the CUDA device
CUDA kernel launch with 196 blocks of 256 threads
Copy output data from the CUDA device to the host memory
Test PASSED
Done
That is, I managed to install CUDA 8.0, which was confirmed by running nvcc --version:
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2016 NVIDIA Corporation
Built on Tue_Jan_10_13:22:03_CST_2017
Cuda compilation tools, release 8.0, V8.0.61
However, unfortunately for me, according to the documentation https://github.com/JuliaGPU/CUDA.jl it seems that CUDA.jl requires CUDA 11.0. However, continuing to read the documentation it seems that it would be possible to use CUDA.jl v1.3 with CUDA 9-10.0. I do not have CUDA 9 for the reason mentioned above.
So, considering my problem, from what I understand so far I will not be able to use CUDA.jl. Is that correct? Or would there still be a way around this?