In preparation for the NCAR provided GPU Training course starting in February, feel free to explore this curated collection of Resources and 3rd Party Learning Materials developed and made available externally. Much of this material will be utilized to inspire learning objectives for the previously mentioned training course but will be adapted to more strongly serve the Earth System Science Community.
The location of this material can be found at this publicly accessible Google Drive folder. If you would like access to a collection of pdf textbooks similarly within this space, please contact Daniel Howard. Only UCAR staff with an active account can access that material as they are only available via our library services.
Some useful trainings/documents to highlight are listed below:
- OLCF CUDA Training Series, 2020
- PRACE Directive Based GPU Computing with OpenACC, 2021
- ENCSS OpenACC/CUDA Training for Beginners, 2020
- NERSC Roofline on NVIDIA GPUs (NSight Compute), 2020
- More recent SC21 Tutorial from NERSC and LBL
- PRACE Best Practices Guide - Modern Accelerators, 2021
- OpenACC Programming and Best Practices Guide, May 2021
- NVIDIA A100 Tensor Core GPU Architecture whitepaper
- High Performance Computing for Applications in Engineering, Dan Negrut UW-Madison Spring 2021
- Gets into the weeds for a semester long course but at least enjoyable to peruse alongside the interwoven comics and jokes.
If there are gaps in material, other resources, or further input you'd like to share about this repository, please reach out and we would be happy to improve upon this resource. As the 2022 training series progresses, we may make time to post in this blog specific GPU usage scenarios and idiosyncrasies relevant to the GPU clusters on Casper and later Derecho. Feel free to add your own material here the same. If this content proves useful, we may further refine it for publication as part of the publicly accessible arc.ucar.edu documentation.