How to use CUDA 11 or Later with FSL 6.0.6.x on Ubuntu 20.04/22.04

FSL 6.0.6 and later now support CUDA 11 or later.
After various trials and errors, I have found a simple way to use CUDA effectively with FSL, which I will introduce here.

Assuming that FSL 6.0.6 or later is already installed.

For those who want to solve it quickly

Please run the following command in the terminal. After that, restart your computer and you’re done.

Setup

• Ubuntu 20.04

cd ~/Downloads
wget https://gitlab.com/kytk/lin4neuro-focal/-/raw/master/installer-scripts/cuda_installer.sh
bash cuda_installer.sh

• Ubuntu 22.04

cd ~/Downloads
wget https://gitlab.com/kytk/lin4neuro-jammy/-/raw/main/installer-scripts/cuda_installer.sh
bash cuda_installer.sh

Verification

After reboot, type the following command.

/usr/local/cuda/bin/nvcc --version

This will display the version of CUDA.

Testing

I have prepared test scripts for each of the following, which will work on both Ubuntu 20.04 and 22.04.

• eddy (about 10 minutes on GPU)

cd ~/Downloads
wget https://gitlab.com/kytk/lin4neuro-focal/-/raw/master/test-scripts/test_eddy_cuda.sh
bash test_eddy_cuda.sh

• xtract (about 40 minutes on GPU; this script was written by Tetsuo Koyama)

cd ~/Downloads
wget https://gitlab.com/kytk/lin4neuro-focal/-/raw/master/test-scripts/test_xtract_gpu.sh
bash test_xtract_gpu.sh

For those who want to test more thoroughly

To be described in the future.