Accelerate AI Training Workloads with Google Cloud TPUs and GPUs

Explore key considerations for choosing tensor processing units (TPUs) and graphics processing units (GPUs) for AI training workloads in this 44-minute session from Google Cloud Next 2024. Learn about the strengths of each accelerator for various workloads, including large language models and generative AI. Discover best practices for optimizing training workflows on Google Cloud using TPUs and GPUs. Understand performance and cost implications, along with strategies for cost optimization at scale. Dive into topics such as accelerated processing kits, HPC toolkits, and the Maloco framework. Gain insights on contextual AI, pretraining ML, and scaling TPUs for high-performance AI training.

Introduction
Key Considerations
Performance
Useful Work
Frameworks
Reference Implementation
Orchestration
Accelerated Processing Kit
HPC Toolkit
Maloco Introduction
How Maloo Works
What Sets Maloo Apart
Maloo Scale
TPUs
Contextual AI
Pretraining ML