Harnessing the Powers of Cloud TPU:
Examples and Walkthroughs
11th October 2018
Google Next 2018, London
Presentation: live slides
Presenters
Thank you
Disclaimer
Agenda
Familiarity
Notebooks?
Google Colab?
GPUs & TPUs
CPU v/s GPU v/s TPU: Simple benchmarking example
Using notebooks on Google Colab
CPU: Simple benchmarking example
CPU: Simple benchmarking example
CPU v/s GPU v/s TPU: Time-series prediction example
Using notebooks on Google Colab
CPU: Time-series prediction example
If it does not work on Google Colab, try running it on your local machine to capture benchmarking stats
GPU: Time-series prediction example
Check out the impressive stats across the entire notebook
TPU: Time-series prediction example
WIP: beat us to it, any contributions or suggestions to help fix the errors, from the community is welcome
Summary
CPU, GPU, TPU: Simple benchmarks results:
TPUs are ~85x to ~312x faster than CPUs
GPUs are ~30x faster than CPUs
also means
TPUs are ~3x to ~10x faster than GPUs
which are ~30x faster than CPUs
Resources
Citations
Some of images used in this presentation are owned by the respective authors, and most of them come from the https://thenounproject.com.
All the notebooks are either from examples provided via Google Colab or from the respective authors, already cited in the notebooks itself.
Thank you, again...
Feedback and contact
Please share your feedback, to be applied to the live slides for everyone’s benefit