Skip to end of metadata
Go to start of metadata

This assessment is targeted towards university-level faculty and students interested in assessing their knowledge of GPUs for HPC. We plan to offer at least two badges pertaining to GPUs for HPC. The first is a GPU Hardware for HPC Badge. This badge is an introductory-level assessment consisting of short quiz and a practical assessment focused on the NVIDIA GPUs currently found in XSEDE systems Expanse, Bridges 2, and Frontera. 

Learning Competencies

Our set of learning competencies were guided in part by the training material available from XSEDE sites such as SDSC, PSC, and TACC, as well as the Cornell Virtual Workshop tutorial on GPU hardware and general information made available from NVIDIA. In particular, the training webinars offered by SDSC may be particularly useful.

In order to pass this assessment, Learners should be able to:

  • Explain what a GPU is and describe how it works.
  • List some specifications for a modern-day GPU used in HPC such as the NVIDIA V100.
  • Summarize why GPUs are used in HPC.
  • Explain how a GPU compares with a CPU, i.e., advantages/disadvantages.
  • Describe what a profiling tool is and how it may be used to evaluate GPU hardware performance. 
  • Demonstrate how to use nvprof for GPU profiling.
  • Illustrate how communication between GPUs and CPUs works.
  • Describe communication between GPUs.
  • Describe the role of NVLink.

Update, April 4, 2021:

The learning competencies were reviewed by Sandie Kappes and Victor Eijkhout.

GPU Hardware for HPC Badge

The GPU Hardware for HPC Badge consists of a relatively simple 10-question quiz made with basic questions about GPU hardware found in GPUs such as the NVIDIA V100 SMX2, and a practical requirement consisting of submitting the output of the nvidia-smi command and the nvprof --gpu_profile_trace command. The quiz requires a 2-hour time limit to complete, and allows up to 3 submissions. 

Update, April 13, 2021: The GPU Hardware for HPC Badge is ready for review. The badge can be found here:

https://www.hpc-training.org/xsede/moodle/course/view.php?id=60

GPU-CUDA for HPC

[TBD by June 2021]


  • No labels