Leveraging the NVIDIA A100 GPU for AI and HPC

Nov 9, 2021, 4:30 pm6:00 pm
Lewis Library 120
Event Description

The NVIDIA A100 GPU will soon be found across the majority of the Research Computing clusters. This powerful accelerator offers a theoretical performance of 9.7 TFLOPS in double precision and 19.5 in single. Specialized hardware units on the GPUs called Tensor Cores theoretically allow for even faster speeds. But in order to take full advantage of the A100, most applications require researchers to modify their input scripts.

This workshop provides an overview of the features of the A100 GPU along with specific use cases for deep learning (PyTorch and TensorFlow) and HPC. Tools for performance profiling and for measuring data transfer rates will be presented.

Workshop format: Lecture and hands-on

Target audience: This workshop is aimed at researchers who are running GPU-enabled codes and would like to learn how best to use the newer A100s in their research.

Learning objectives: Attendees will learn about the new features and capabilities of the A100 GPU and come away with the ability to modify their application scripts and SLURM scripts to take full advantage of these new GPUs.

Knowledge prerequisites: Users should be familiar with either a deep learning framework (e.g., PyTorch, TensorFlow) or a GPU-enabled research code.

Hardware/software prerequisites: For this workshop, users must have an account on the Adroit cluster, and they should confirm that they can SSH into Della/Adroit *at least 48 hours beforehand* (users who have existing accounts on DellaGPU can use that for the workshop, but some examples may run more slowly if they are competing with higher priority jobs). Details on all of the above can be found in this guide. THERE WILL BE LITTLE TO NO TROUBLESHOOTING DURING THE WORKSHOP!