{"id":"arxiv/2609.12307","url":"https://arxiv.org/abs/2609.12307","tags":["math.NA"],"score":0,"title":"How to grade the accuracy of the BLAS","domain":"math.NA","source":"arxiv","authors":["James Demmel","Greg Henry","Igor Kozachenko","Julien Langou","Xiaoye Sherry Li","Jason Riedy","Jackson Vanover"],"reviews":0,"subject":"math.NA","summary":"Motivated by accelerating machine learning (ML), many computer vendors and chip manufacturers are building accelerators for matrix multiplication, which save time and energy by operating in the lower precisions needed for ML. This has in turn motivated many efforts to use these accelerators to provide faster matrix multiplication implementations with the higher precision required by many other linear algebra applications. Motivated by the large design space of algorithms for approximating higher precision, with significant performance/accuracy tradeoffs, we provide a benchmark to \"grade the…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-09-11","reactions":{}}