3 ms·
> some global technique to solve PCA (SVD), which cannot take advantage of the parallelism of GPU / TPU My reading of https://developer.download.nvidia.com/vid
by bigbillheck 5y ago
> some global technique to solve PCA (SVD), which cannot take advantage of the parallelism of GPU / TPU
My reading of https://developer.download.nvidia.com/video/gputechconf/gtc/2019/presentation/s9226-fast-singular-value-decomposition-on-gpus-v2.pdf https://developer.download.nvidia.com/video/gputechconf/gtc/... suggests that it's just the usual modern SVD algorithm, in particular the QR factorization part, that's the limiting factor, and with some thought there are ways to do better.