Abstract
Concurrent application demands that exceed available GPU resources can degrade performance and worsen the user experience. However, there are no easy mechanisms for lay users to tailor application performance and control GPU resource allocation. This disclosure describes techniques that enable users to specify preferences for prioritizing computational tasks performed via a GPU via conventional user interface mechanisms. Scheduling and switching workloads and batches of the requested computations on the GPU is performed according to user-specified preferences. Users can leverage the preference settings for managing performance tradeoffs between application functionality, balancing resource allocation when multitasking, troubleshooting potential GPU-related problems, etc. The techniques can enable users to save time and effort in troubleshooting GPU-related problems and can save manufacturers the cost of handling return and replacement of defective hardware.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Schneider, Chris and Watson, Nic, "Mechanism for Balancing of GPU Usage Between AI Queue Depth and Graphical Fidelity", Technical Disclosure Commons, (March 13, 2023)
https://www.tdcommons.org/dpubs_series/5737