Skip to content

Comment on The Nvidia DGX-1 Deep Learning Supercomputer in a Boxparent

Comments

It doesn't look like AMD's HIP effort compiles to OpenCL, but to C++, that works on specific AMD GPUs (Fiji R9 Fury X, R9 Fury, R9 Nano).

Also, much of the speed gains for ML on NVidia hardware come from CuDNN - there is no equivalent for OpenCL or AMD hardware

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.