As we will use GPUs in the assignments and your final project, please check your account and let us know if you have any question.
There are two ways to access the GPU server: (1) Datahub and (2) ssh command line. First option is easier to set up so we recommend starting with the first option.
You might still want to access the server via ssh if you want advanced use cases such as file transfer.
1st option: Datahub (easier); courtesy of COGS 181 instructional team
You can also use Datahub/Jupyterhub (https://datahub.ucsd.edu/), which provides access to web-based Jupyter notebooks.
1. Go to https://datahub.ucsd.edu/ and click the yellow button to login with your UCSD account.
2. Select the available environment. You may want to choose the one with GPU.
3. You will enter the jupyter notebook with the environment of pytorch. You can now start your HW by uploading the file or creating a new one.
4. When you finish, remember to save your work and click Control Panel at upper right to "Stop My Service" and logout.
2nd option: SSH command line (advanced)
For this class, we have summarized what is important to know just to get you started here in GPU_usage_guide.pdf. Below are the most important steps.
To login to your account, you need first reset the password through:
(https://acms.ucsd.edu/students/gpasswd.html )
Then open the terminal and type:
ssh ACCOUNT@ieng6.ucsd.edu
Password:
launch-scipy-ml-gpu.sh
nvidia-smi
If you find some error such as "command not found" after typing the "launch-scipy-ml-gpu.sh", you may need to use command "prep" at first, and run the script "launch-scipy-ml-gpu.sh"
Please log in the server using UCSD internet access.
3. You should be able to see the status of the GPU that you are currently assigned. By copy-pasting the URL to your local web browser, you should be able to connect to the remote notebook server.
The complete guide is available here:
The course configured with the "scipy-ml" image (it includes PyTorch, TensorFlow, and general CUDA support), as well as per-student limits of 4 CPU, 16 GB RAM, and 1 GPU. In general, you should not need to install any dependencies. If you are trying something new and really need to install your own dependencies, please follow the most up to date tutorial here (how to set up docker).