Hands-on labs will be done using VSC-5 – Austria's flagship supercomputer.
Login for the participants will be possible (unlocked)
from Tuesday, 24 October 2023, 08:30 CEST to Wednesday, 25 October 2023, 19:00 CEST.
GPUs (node allocation) for the participants will be available
when the first hands-on lab starts at 10:00 CEST.
We will login (ssh) with a shared user (trainee99) and provide the password in Zoom.
There will be 20 GPU nodes each with 2 GPUs (NVIDIA A100) reserved for the course,
i.e., 2 participants share 1 node and each participant gets interactive access to 1 (own) GPU.
Please take care to not interfere with the other participants!
VSC login for the CUDA course
1. Start your local X (if available)!
2. Login to vsc5:
ssh -tX trainee99@vmos.vsc.ac.at vsc5
You'll be asked for your password twice (on vmos and on vsc5):
The password will be provided in Zoom.
2.a. PuTTY settings – only for those who use Windows together with PuTTY:
If you use PuTTY please enter (see screenshot):
HostName: vmos.vsc.ac.at
Port: 22
Connection type: SSH
Left menu --> SSH --> Remote command: vsc5
Open --> in terminal - login as: trainee99
You'll be asked for your password twice (vmos and vsc5):
Note, the password will be invisible when you type it.
3. When you are logged in to vsc5, figure out your node, ID (working directory), and GPU:
grep -i yourname PARTICIPANTS_GPU
e.g. for tony: grep -i tony PARTICIPANTS_GPU
Tony Testing ssh -X n3071-002 cd ~/02 export CUDA_VISIBLE_DEVICES=1
4. Use your node, ID (working directory), and GPU (copy from above output):
4.a. ssh command to access your node
ssh -X n...
4.b. cd command to your own directory
cd ~/..
4.c. switch to use your own GPU
export CUDA_VISIBLE...
4.d. spack-load & setup cuda@11.6.2
source ~/cuda_setup.sh
