Overview
Watch a technical presentation from Google Cloud's storage group that explores their comprehensive AI infrastructure optimization efforts. Dive into the AI hypercomputer system that combines advanced hardware accelerators like GPUs and TPUs with robust software solutions. Learn about new storage innovations including Parallel Store, Google Cloud's first managed parallel file system, and Hyperdisk ML, a read-only block storage service designed specifically for AI workloads. Understand how different storage solutions cater to various scales of AI operations, from small-scale low latency jobs to large-scale high-throughput needs. Discover Google Cloud's open ecosystem partnerships and their approach to providing consistent, flexible storage solutions that bridge on-premises and cloud environments. Gain insights into selecting appropriate storage solutions based on training data sets, checkpointing requirements, and serving needs for optimal AI infrastructure management.
Syllabus
Workload and AI-Optimized Infrastructure from Google Cloud
Taught by
Tech Field Day