Programming ML Supercomputers: A Deep Dive on Cloud TPUs (Next Rewind ’18)


Recent increases in computational power have allowed deep learning techniques to achieve breakthroughs on previously intractable problems including image understanding, translation, game playing (Go, Atari, etc.), voice interactions, and more. In order to power the next generation of products, Google Cloud developed Cloud TPUs, which allow more advanced deep learning models to be trained on larger datasets faster and more cost effectively than ever before. This talk will be a technical deep dive on when to use Cloud TPUs, as well as how to program them. We will cover the programming abstractions that allow you to run your models on CPUs, GPUs, and Cloud TPU, from single devices up to entire Cloud TPU pods.

Original talk by Brennan Saeta
Rewind by Yufeng Guo

Watch full session here → http://bit.ly/2DwdD5r
Watch more recaps here → http://bit.ly/NextRewind2018

Watch more Infrastructure & Operations sessions here → http://bit.ly/2uEykpQ
Next ‘18 All Sessions playlist → http://bit.ly/Allsessions

Subscribe to the Google Cloud Platform channel! → http://bit.ly/GCloudPlatform


Duration: 3:42
Publisher: Google Cloud
You can watch this video also at the source.