2026-07-03

2 saved articles
BackLibrary
ArticleRead
js13kgames.com
js13kgames.com / source link only
--
ParallelKernelBench: Frontier LLMs can't write fast multi-GPU kernels (yet)
Willy Chan, Nathan Paek, Simon Guo, Simran Arora, Daniel Y. Fu / www.together.ai
ParallelKernelBench tests whether LLMs can write fast multi-GPU CUDA kernels across 87 real workloads. The best model solves under a third, but a few generated kernels beat any public implementation.
--