Practical AI Going full bore with Graphcore!
Apr 13, 2021
Dave Lacey, Chief Software Architect at Graphcore, dives into the world of AI by connecting popular user interfaces like TensorFlow and PyTorch to cutting-edge hardware capabilities. He explains the significance of the Poplar Graph Framework Software and how it revolutionizes AI development. The conversation touches on innovative co-design between hardware and software, mechanical sympathy in coding, and the need for adaptive strategies as AI technology evolves. Lacey's insights make the complex landscape of AI hardware accessible and engaging.
AI Snips
Chapters
Transcript
Episode notes
AI Hardware Landscape
- AI hardware landscape includes CPUs, GPUs, and specialized processors like Graphcore's IPU.
- CPUs excel at scalar processing, GPUs at vector processing, and IPUs at graph processing, handling interconnected data.
Processor Categorization
- Categorize processors by data type and processing style.
- CPUs handle scalar data, GPUs large vectors, and IPUs interconnected graph data, reflecting varying AI workload needs.
Graph Processing in AI
- Graph processors efficiently handle interconnected data, unlike traditional vector processing.
- Convolutional operations and newer neural network architectures like ResNeXt utilize this graph-like data flow.