GPU Compute facility Details

First Time Access

STEP 1: (for first time access only) visit the URL "https://gpu.iitg.ac.in/":

  1. click on "Forgot Password" to reset your password.
  2. For the input "Username or Email", type in your (without the @iitg.ac.in) OR .
  3. For the input "Captcha", type in the exact case sensitive string displayed.
  4. click on "Send OTP".
Note: An OTP will be sent to your IITG Email.

STEP 2: (only after STEP 1 was successfully completed):

  1. For the input "OTP Code", enter the OTP received on your IITG Email. OTP valid for 600 seconds only.
  2. For the input "Captcha", type in the exact case sensitive string displayed.
  3. Click on "Verify OTP".
Note: You will need to type in your choice of password in the next screen.

STEP 3: (only after STEP 3 was successfully completed):

  1. For the input "Captcha", type in the exact case sensitive string displayed.
  2. For the input "New Password", type in the password of your choice.
  3. For the input "Confirm New Password", re-type the exact password chosen.
  4. Click on "Reset Password".
  5. Click "Ok" on the Password has been reset successfully, Alert.
Note: You GPU Compute login password has now been set to the password chosen by you.

STEP 4: (only after you have confirmed the availability of a valid credential to access the GPU Compute Facility)

  1. From within a Terminal Application (Linux Terminal, Windows cmd, putty, etc.) type the following:
  2. ssh <username>@gpu.iitg.ac.in
    where, <username> will be your IITG Email ID (without the @iitg.ac.in)
  3. Enter your password at the prompt.
    If the entered credentials are correct, you will be presented with a similar prompt below:
    <username>@gpu:~$
Note: You have now access to the GPU Compute facility.

IIT Guwahati GPU Cluster User Manual

1. Introduction

This guide provides the basic steps required to access the IIT Guwahati GPU Cluster, prepare your working environment, submit GPU jobs using slurm, and monitor job execution.

2. Logging In
ssh <username>@gpu.iitg.ac.in

Use your assigned username and password. Windows users may use PuTTY; Linux and macOS users can use the built-in SSH client.

3. Password Management

To change or reset your password, visit https://gpu.iitg.ac.in, select the options available and follow the instructions.

4. Working Directories

Home directory (/home/<your_username>): Store source code, scripts, and small files only.

Scratch directory (/scratch/<your_username>): Should be used for input datasets, temporary files, and job execution.

Example:
cd /scratch/<your_username> mkdir my_project
cd my_project
5. Creating and Submitting a Job

Create a file "gpu_job.sh" with the contents below (provided as a template), in your <my_projects> folder

#!/bin/bash
#SBATCH --job-name=gpu_test
#SBATCH --partition=gpu_small
#SBATCH --ntasks=1
#SBATCH --gres=gpu:1
#SBATCH --output=gpu_test_%j.out
#SBATCH --error=gpu_test_%j.err
#SBATCH --mail-user=your-email@example.com
#SBATCH --mail-type=ALL
echo "Job started at $(date)" hostname
nvidia-smi
# Add your application below
...
echo "Finished at $(date)"

Submit from your scratch project directory

sbatch gpu_job.sh
6. Monitoring Jobs
Check Cluster Status 
sinfo
View Jobs 
squeue
squeue -u $USER
scancel
Note: Output files (*.out and *.err) are created in the directory where the job was submitted.
7. Best Practices
  • Always execute jobs from /scratch/<your_username>/.
  • Do not run computational workloads on the login node.
  • Request only the CPUs, memory, GPUs and Walltime required.
  • Delete unnecessary files from your scratch partition periodically.
  • Monitor jobs regularly using squeue.
  • Cancel jobs that are no longer required.
8. Troubleshooting

Permission denied : Verify username/password and SSH access

Job pending : Requested resources may not yet be available

GPU not detected : Ensure you requested GPUs using --gres=gpu:1

Support

For account-related issues, password reset, quota, or job scheduling assistance, contact the GPU Cluster Administration Team (hpcadmin@iitg.ac.in).