CERN Computing Seminar

Graphcore Technical Overview: Designing Hardware and Software for Next Gen Machine Intelligence Models

by Alex Titterton (Graphcore)

Europe/Zurich
Zoom Webinar: (https://cern.zoom.us/j/92250037748)

Zoom Webinar:

https://cern.zoom.us/j/92250037748

https://cern.zoom.us/j/92250037748
Description

As machine learning workloads explode in complexity and size, new hardware and software is required to support next gen AI models for both training and +inference. This talk will highlight the evolution of Machine Intelligence, why Graphcore invented the IPU and go into detail on the co-designed Poplar Software Stack – the +world’s first graph tool chain. It will also summarise our performance benchmarks and how the IPU enables cutting-edge applications such as ResNext.

About the speaker

Growing up, Alex Titterton became interested primarily in mathematics, computers, coffee and old Italian cars, and moved south in order to undertake a master’s +degree in mathematics and physics at the University of Bristol. In 2015 he started a PhD in particle physics, jointly appointed between the University of Bristol, University +of Southampton and the Rutherford Appleton Laboratory, and as part of this in 2017 moved to France to work at CERN. For 18 months he spent the weekdays searching for +Supersymmetry and the weekends skiing and driving an old Alfa Romeo around scenic mountain roads. A thesis later, he finally finished his PhD in Bristol, and found a new home +at Graphcore, working first in the Product Support Engineering team and most recently moved to a new role as an AI Field Engineer.

Zoom Webinar: 92250037748

More information
Organised by

Ricardo Brito Da Rocha, - IT Department


NOTE: In order to prevent spurious disturbances, participants joining via Zoom are kindly invited to keep their camera and microphone muted until the Q&A part after the presentation. When it will be your turn, enable to pose your comments/questions, then mute again