CURC's largest supercomputer, Alpine, offers users access to state-of-the-art hardware, including AMD MI100 GPUs and NVIDIA A100 GPUs. Though this hardware can process complex deep learning workflows and significantly reduce run times for some computations, users must first overcome the barriers to entry before taking full advantage.
This Primer*, aimed at researchers with little to no GPU programming experience, introduces heterogeneous (CPU/GPU) computing and popular GPU programming models. The session also presents criteria for porting workflows to GPUs and a high-level overview of factors that determine speed-up. Participants will learn how to request GPU nodes on Alpine with Slurm, then try a hands-on example to compare run times for CPU-only and GPU-accelerated scripts. Examples will focus on Python and C programming languages, but experience with either language is not required.
Participants should have a personal computer and a Research Computing Account prior to the start of the training if they wish to try hands-on examples.
*What is a Primer? A Primer is a session that provides an introduction to concepts, systems, and tools. You can expect to leave a Primer with an introductory-level understanding of the topic.