Reproduce “Derived, Order-Robust Tolerances for Verifying Nondeterministic GPU Reductions” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Vadhyar Sai Charan, Bora Sai Sathvik
The authors derive an input-dependent tolerance for comparing two GPU reductions without calibration, using probabilistic rounding-error analysis and an order-robust refinement. Across 145 nondeterministic CPU and GPU cells, none exceeded tested tolerances; maximum utilization was 36% on CPU and 29% on GPUs. This observed coverage does not confirm the designed failure probability, and the guarantee excludes measured biased paths.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.