High Performance Computing on Google TPUs
July 28, 2026
Online
Embedded Files
About the Session
Serving as the platform for numerous AI models across training and inference, Google's Tensor Processing Units (TPUs) offer immense capabilities beyond standard artificial intelligence tasks. Due to their innovative architecture and extreme scalability, they unlock new possibilities for alternative high-performance computing (HPC) workloads. This session, featuring a hands-on Colab demonstration, explores how traditional HPC challenges can be effectively addressed by leveraging TPUs.
The session will begin with an examination of TPU TensorCore capabilities, network design, and mathematical precision. Following this, we will explore case studies of conventional high-performance computing applications executed on TPUs. The presentation will conclude with an interactive demonstration, allowing attendees to experiment firsthand via a Google Colab notebook focused on distributed fast Fourier transforms.
This presentation offers a practical blueprint and conceptual framework for utilizing TPUs as a modern accelerator architecture across both HPC and AI workloads, tailored for research engineers, data scientists, and HPC practitioners alike.
Speaker
John Fonner
Product Manager, Google Cloud
As a Product Manager in the High Performance Computing group in Google Cloud, John Fonner is dedicated to making Google's planet-scale computing infrastructure the premier platform for public sector research. With nearly two decades of experience in scientific high-performance computing and a background in Biomedical Engineering, John has a proven track record of leading and contributing to large-scale cyberinfrastructure projects that broaden the reach of scientific resources and accelerate the pace of discovery. Prior to joining Google, John was the Director of Special Projects at the Texas Advanced Computing Center, where he led and contributed to numerous research and cyberinfrastructure projects funded by the NSF, NIH, DARPA, and others.
Google Sites
Report abuse