Graphcore dives deep into software architecture layers on Practical AI Podcast

Engineers are increasingly looking for greater choice in how to process the variety of data structures involved in modern machine learning models. Having a flexible software stack not only helps with these choices but makes it easier to pursue the development of new types of models.

During a recent episode of the Practical AI Podcast, Graphcore Chief Software Architect Dave Lacey spoke about what makes AI data different, why co-designing software and hardware matters and how the many different layers of the Poplar software stack communicate with the IPU.

Practical AI 129: Going full bore with Graphcore! – Listen on Changelog.com