Accelerator-Level Parallelism (ALP)


M. Hill and V. J. Reddi, “Accelerator-Level Parallelism (ALP),” arXiv, vol. arXiv:1907.02064 [cs.DC], 2019.
Paper.pdf1.02 MB


With the slowing of technology scaling, the only known way to further improve computer system performance under energy constraints is to employ hardware ​accelerators​. Already today, many chips in m​ obile, edge and cloud computing concurrently employ multiple accelerators in what we call ​accelerator-level parallelism (ALP)​. For the needed benefits of ALP to spread to computer systems more broadly, we herein charge the community to develop better “best practices” for: targeting accelerators, managing accelerator concurrency, choreographing inter-accelerator communication, and productively programming accelerators.

Last updated on 10/09/2020