Guarding Numerics Amidst Rising Heterogeneity
Reliability and Resiliency
TimeFriday, 19 November 20219am - 9:20am CST
DescriptionNew heterogeneous computing platforms---especially GPUs and other accelerators---are being introduced at a brisk pace, motivated by the goals of exploiting parallelism and reducing data movement. Unfortunately, their sheer variety as well as the optimization options supported by them have been observed to alter the computed numerical results to the extent that reproducible results are no longer possible to obtain without extra effort. Our main contribution in this paper is to document the scope and magnitude of this problem that has not received much attention. We propose a taxonomy of sub-topics and details under each, and close with a summary table of these discussions. Given the magnitude of maintaining reliable information, we issue a call for community action, helping us maintain a website containing curated information.