Havvarsel

For the researchers in the Havvarsel project, access could be granted to the GPU nodes of MET. Follow the subsequent instructions:

Requirements

• User account at MET
• Google Authenticator App
• Pritunl Client on local machine

Connecting

1. Start Pritunl Client using the given PIN and the one-time password (OTP) from Google Authenticator App
2. Further documentation can be found on dokit.met.no or especially PPI
3. Connect to the GPU nodes in two steps. First, to VGL servers at MET using the LDAP password

$ ssh <username>@vglserver2.met.no

and then to the GPU node using the LDAP password

$ ssh <username>@gpu-0<number>.ppi.met.no

Setting up a run

1. The CUDA drivers are already installed, but the CUDA toolkit needs to be installed for Linux x86_64 CentOS 7 using the runfile (local):

$ wget https://developer.download.nvidia.com/compute/cuda/11.2.0/local_installers/cuda_11.2.0_460.27.04_linux.run
$ chmod +x cuda_11.2.0_460.27.04_linux.run
$ ./cuda_11.2.0_460.27.04_linux.run

In the installation menu only the CUDA toolkit should be selected (no CUDA drivers!) and installed to a folder with writing permission, e.g. /home/<username>/cuda. The respective location has to be added to PATH, e.g. in .bashrc:

export PATH=$PATH:/home/<username>/cuda/bin

2. Follow the instructions in README
3. Run jobs in the background. Start with

$ nice --0 screen
$ screen -S <screenname>