Published on

Go Faster with float16


Sam Witteveen and I started the TensorFlow and Deep Learning Singapore group on MeetUp in February 2017, and the seventeenth MeetUp, aka TensorFlow and Deep Learning: Go Faster!, was again hosted by Google Singapore.

To some extent the theme of the MeetUp was set by Jason Zaman volunteering to talk about "Optimising your TensorFlow installation", based on his extensive experience as the TensorFlow maintainer for the Gentoo project. Sam Witteveen (via video hook-up from New York) talked about the new Deep Learning VM images that Google released less than 2 weeks ago. We also had a lightning talk by Hardeep Arora about doing face detection in the browser using TensorFlow.js.

For my part, I gave a talk titled "Go Faster with float16", which explained some of the intricacies in training a model in a mixed-precision environment, such as might be used on a P100 or V100 GPU (if you want to use the float16 cores - or even the TensorCores - provided by these server-class Nvidia chips).

The slides for my talk are here :

Presentation Screenshot

If there are any questions about the presentation please ask below, or contact me using the details given on the slides themselves.

Presentation Content Example