Generalized Resource Estimation

Overview

Welcome! This guide is intended for new researchers interested in submitting a NAIRR Pilot proposal and has little experience with HPC or large-scale resources.

Learning Outcomes

  • Translate a research task into CPU-hours, GPU-hours, node-hours, RAM and scratch & long-term storage.

  • Produce a back-of-the-envelope estimate from small pilot timings.

Key Words

  • CPU core-hour = 1 CPU core used for 1 hour

  • GPU-hour = 1 GPU used for 1 hour

  • GB-hour = memory in GB × hours used

  • TB-month = storage used averaged over a month

  • SU (Service Unit) = accounting unit that maps to time on a specific resource

An example of SU units: Jetstream2 GPU virtual machines (VMs) cost 2x vCPUs. G3.large VM costs 32 SUs per hour. NAIRR pilot can request 128,000 SUs (2000 GPU-hours)

NAIRR

This material resource is to guide you to estimate the compute resources you might need while requesting HPC or large-scale resources from National Artificial Intelligence Research Resource (NAIRR) Pilot program.

NAIRR Pilot connects U.S. researchers and educators to computational, data, and training resources needed to advance AI research. This program works through shared Federal and private compute infrastructure to birdge the gap between researchers and educators to AI resources. More information reagaring NAIRR and how to acknowledge them can be found at: https://nairrpilot.org/about.

The information regarding which all resources are available for researchers can be found at: https://nairrpilot.org/opportunities/allocations and the same for educators can be found at https://nairrpilot.org/opportunities/education-call. A detailed video on instructions to request for resources is published in a youtube video at: https://www.youtube.com/watch?v=GCTv5OjI1ys&t=184s, with link to the slide used shared in the description.

Add your content using reStructuredText syntax. See the reStructuredText documentation for details.