Using the GPU partition for GPU-Intensive Jobs

Created by Bent Petersen, Modified on Mon, 3 Mar at 4:02 PM by Bent Petersen

The gpuqueue partition is designed for GPU-based computations, such as machine learning, deep learning, and GPU-accelerated bioinformatics tools.

Key Features

• Max Runtime: Up to 14 days (336 hours).

• Available GPUs: NVIDIA A100 GPUs (up to 4 per node).

• Max CPU Allocation: Subject to available GPUs per job.

• Memory Limit: Subject to available GPU and node memory.

• Priority: Standard fair-share based scheduling.

• Best for: TensorFlow, PyTorch, CUDA, GPU-accelerated computations.

How to Submit a Job to the GPU Queue

To submit a GPU job, specify the gpuqueue partition and request GPU resources:

sbatch --partition=gpuqueue --qos=normal --gres=gpu:1 --cpus-per-task=8 --mem=64G --time=12:00:00 --wrap="your_gpu_command_here"

Best Practices

• Always request GPUs with --gres=gpu:N (replace N with 1-4 as needed).

• Ensure your application supports GPU acceleration before using this queue.

• Do not use the GPU queue for CPU-only jobs; they belong in cpuqueue.