Demystifying the Placement Policies of the NVIDIA GPU Thread Block Scheduler for Concurrent Kernels
ACM SIGMETRICS Performance Evaluation Review · 05.03.2021 · 48 (3) · 81-88
DOI: 10.1145/3453953.3453972 ↗Kurz gefasst
Fachartikel in der Zeitschrift ACM SIGMETRICS Performance Evaluation Review, online erschienen am 05.03.2021. Verfasst von Guin Gilman und 3 weiteren Autor:innen. Die Arbeit wurde bisher 31-mal zitiert (Crossref) und nennt 20 Literaturangaben.
Abstract (Originalsprache)
In this work, we empirically derive the scheduler's behavior under concurrent workloads for NVIDIA's Pascal, Volta, and Turing microarchitectures. In contrast to past studies that suggest the scheduler uses a round-robin policy to assign thread blocks to streaming multiprocessors (SMs), we instead find that the scheduler chooses the next SM based on the SM's local resource availability. We show how this scheduling policy can lead to significant, and seemingly counter-intuitive, performance degradation; for example, a decrease of one thread per block resulted in a 3.58X increase in execution time for one kernel in our experiments. We hope that our work will be useful for improving the accuracy of GPU simulators and aid in the development of novel scheduling algorithms.
Zitieren
APA
Gilman, G., Ogden, S. S., Guo, T., & Walls, R. J. (2021). Demystifying the Placement Policies of the NVIDIA GPU Thread Block Scheduler for Concurrent Kernels. ACM SIGMETRICS Performance Evaluation Review, 48(3), 81-88. https://doi.org/10.1145/3453953.3453972
Vancouver
Gilman G, Ogden SS, Guo T, Walls RJ. Demystifying the Placement Policies of the NVIDIA GPU Thread Block Scheduler for Concurrent Kernels. ACM SIGMETRICS Performance Evaluation Review. 2021;48(3):81-88. doi:10.1145/3453953.3453972
BibTeX
@article{Gilman2021,
author = {Gilman, Guin and Ogden, Samuel S. and Guo, Tian and Walls, Robert J.},
title = {Demystifying the Placement Policies of the NVIDIA GPU Thread Block Scheduler for Concurrent Kernels},
journal = {ACM SIGMETRICS Performance Evaluation Review},
year = {2021},
volume = {48},
number = {3},
pages = {81--88},
doi = {10.1145/3453953.3453972}
}
EUVIMED