Submission #8: Understanding how to implement sparse matrix operations efficiently ================================================================================== Abstract -------- Lots of modern software is built around linear algebra, for scientific computing, machine learning, AI, or realistic immersive graphics. But we are increasingly encountering /sparse/ matrices which require completely different technology than the GPU acceleration that has been applied so successfully to dense matrices. In this talk we will explore the challenges involved in making efficient the inherently irregular calculations and storage involved in manipulating sparse matrices. This involves addressing several systems problems simultaneously in the relationships between programs, cores, caches, and memory. Author ------ Simon Dobson (University of St Andrews)