Maximizing GPU usage with MIGs, MPS, and time-Slicing [online, CIP203]
Note: all times are shown in the timezone in which each event occurs.
Date: 25 November 2025 @ 09:00 - 12:00
Timezone: Eastern Daylight Time
Language of instruction: English
Registration:: https://www.eventbrite.ca/e/1635687036959
Divide your GPUs to multiply your productivity!
Unlock the full potential of your GPU resources! Learn resource-sharing techniques to transform how you use your GPUs. Maximize your computational efficiency through the use of fractional GPUs or by sharing resources between multiple jobs and tasks. This hands-on workshop is particularly essential for users of powerful H100 systems, as poor GPU utilization leads to wasted resources, lower job priority, and potential job termination.
Plan:
- Why bother maximizing GPU usage
- How to waste GPU cycles
- CUDA streams
- Wasting less GPU resources with the MIGs
- Sharing one GPU between multiple tasks with the MPS
- TimeSlicing
- Lots of exercises (executing jobs on the cluster, running tests, measuring execution times, etc)
Keywords: GPU, HPC
Activity log
