CUDA (General)

Step 1: Prepare CUDA Program Source Code

Example source code path -> /home/$USER/job_template/C/cuda_hello_world.cu

#include <stdio.h>
#include <cuda_runtime.h>

// CUDA kernel to print Hello World
__global__ void helloWorldKernel() {
  printf("Hello World from thread %d, block %d\n", threadIdx.x, blockIdx.x);
}

int main() {
  // Launch the kernel with 1 block of 10 threads
  helloWorldKernel<<<1, 10>>>();

  // Synchronize to ensure all printf calls are completed
  cudaDeviceSynchronize();

  // Check for any CUDA errors
  cudaError_t err = cudaGetLastError();
  if (err != cudaSuccess) {
    fprintf(stderr, "CUDA Error: %s\n", cudaGetErrorString(err));
    return 1;
  }

  return 0;
}

Step 2: Prepare job template script

Pre-configured template script path -> /home/$USER/job_template/slurm_job

#!/bin/bash
#SBATCH --job-name=cuda_hello_world ## Job Name
#SBATCH --partition=shared_gpu_l40 ## Partition for Running Job
#SBATCH --nodes=1 ## Number of Compute Node
#SBATCH --ntasks-per-node=1 # Number of Task per Compute Node
#SBATCH --cpus-per-task=2 ## Number of CPU per task
#SBATCH --time=60:00 ## Job Time Limit (i.e. 60 Minutes)
#SBATCH --mem=10GB ## Total Memory for Job
#SBATCH --gres=gpu:l40:2 # Number of GPUs per node (i.e. 1 x L40 GPU)
#SBATCH --output=./%x%j.out ## Output File Path
#SBATCH --error=./%x%j.err ## Error Log Path

## Initiate Environment Module
source /usr/share/modules/init/profile.sh

## Reset the Environment Module components
module purge

## Load Module
module load cuda/12.6

## Run user command
nvcc -o ./cuda_hw /home/${USER}/job_template/C/cuda_hello_world.cu
./cuda_hw

## Clean up
rm cuda_hw

## Clear Environment Module Components
module purge

Step 3: Create Template (Web Interface Feature)

To submit HPC via web interface a job template is required, details please refer to: Create Template (Web Interface Feature)

For Job submission via CLI Terminal, please skip this step.

Step 4: Submit HPC Job

Guides for submitting HPC job, please refer to: HPC Job Submission