UCI-ICS-GPU

Tutorial on using UCI SLURM Servers for GPU-accelerated tasks

Requirements

• ICS Openlab Account
• Access to UCI SLURM servers

---

Step 1: Log into ICS Openlab Account

1. Log in to your ICS Openlab account.
2. Verify the correct user by running:

   whoami

   This should return your UCI Net ID.

---

Step 2: Submit a Job to UCI SLURM Servers with GPUs

1. Check out the available SLURM servers here:
   UCI SLURM Wiki
2. Load SLURM onto your Openlab instance by running the following commands:

   module load slurm
   module initadd slurm

---

Step 3: Set Up the Environment

1. Clone this repository or create your project. For this tutorial, we will use the Python file included in this repository.
2. Load the Python environment:
   To submit the enviorment load script, use the following command:

sbatch -p opengpu.p load.sh

Running this on the opengpu server will ensure that we correctly load all binary Output of this command will be located in a file called "slurm-{jobnumber}.out" where the job number is of the job you just loaded

Step 4: Running the SLURM Job

We will use an SBATCH script to submit the job. Keep in mind:

• SBATCH "comments" (lines starting with #SBATCH) are not just comments; they configure the nodes assigned by the SLURM instance.
• Detailed documentation can be found here.

To submit the job, use the following command:

sbatch -p opengpu.p slurmscript.sh

Replace slurmscript.sh with the name of your SLURM script or Python file.

---

Step 5: Viewing Output and Errors

To check the output and errors generated by the SLURM script:

• View output logs:

  tail -n +1 -f myoutput.out

• View error logs:

  tail -n +1 -f myerrors.err

---

Additional Resources

• UCI SLURM Wiki