The CSE department has a Linux GPU server offering access to two NVIDA A100 GPUs. Be mindful of application usage. Abuse and misuse of gpu resources will result in a warning, and possibly your access being revoked. For access, send a request to the sysadmin team.
To request access, send an email to cshelp@nmt.edu with your name and reason for usage.
The gpu server can accessed via SSH using the hostname gpu.cs.nmt.edu. Use your CSE account credentials to login.
ssh <username>@gpu.cs.nmt.edu.An SSH server can present a PKI public key to identify itself. Upon first connecting to a server, the ssh client will warn you that the authenticity of a host can't be established. This message is normal and allows you to verify the server. It safe to connect if the ED25519 key fingerprint matches SHA256:rM7X2//359pbzQsQMAYoOU5SgF6R0VVAaXurNnuoNfs. If it does not match, please notify a sysadmin.
conda has been installed for easy setup and management of environments and dependencies. To create a conda environment, run the command conda create --name tf tensorflow-gpu, and navigate the prompts. This will also install the tensorflow-gpu package suite. You may want to run conda init to setup the default conda environment.
To activate the environment, run conda activate <envname>, so in our case, conda activate tf.
It's a good idea to make sure that the GPUs on the server are properly set up and available for use. To do this, start a python session with python.
In the python environment, paste the following diagnostic script:
import tensorflow as tf
gpus = tf.config.list_physical_devices('GPU')
if gpus:
print("Available GPUs:")
for gpu in gpus:
print(gpu.name)
else:
print("No GPUs found.")
If the GPUs are accessible, the script should print:
Available GPUs:
/physical_device:GPU:0
/physical_device:GPU:1
If the script prints something else, including listing more or fewer devices or the message No GPUs found., please consult the sysadmin team for assistance or troubleshooting.