The NAS Application and Performance Productivity (APP) team will conduct Cabeus user training sessions to be delivered in two separate Teams meetings scheduled for March 12 and 13 at 11:00 a.m. Pacific Time.

Part 1 gives an overview of Cabeus - the new GPU cluster at NAS for HPC and AI/ML applications. This session will provide information on accessing Cabeus; describe its hardware resources; define a virtual node and illustrate its effect on PBS jobs sharing; recommend a proper way to use some of the host memory as a temporary filesystem for I/O; and help you understand and check how each job is charged against your SBU allocation.

Part 2 introduces topics relevant to programming and building HPC applications for running on one Nvidia GPU. This session gives a high-level view of a few programming methods recommended by Nvidia, namely CUDA, OpenACC, OpenMP and standard languages; weaves these programming methods with two important concepts including CPU offloading work to GPU and explicit vs. implicit data movement between CPU and GPU; and highlights the roles of compute capability, CUDA toolkit and CUDA driver for building your application.

For details on joining these sessions, see HECC User Training.