Designing Flexible, Efficient AI Infrastructure with Google Cloud's AI Hypercomputer

​As AI models grow in complexity and scale, ML engineering teams need infrastructure that delivers maximum performance without giving up agility or cost efficiency.

​In this AlphaSignal session, Chelsie Czop from Google Cloud walks through how to implement the Design for Flexibility and Efficiency (DFE) framework with NVIDIA AI infrastructure across the AI Hypercomputer architecture. You'll see how to optimize training and inference pipelines across Google Cloud's NVIDIA GPU portfolio. You'll also see how to route workloads dynamically across instance types with GKE Custom Compute Classes and the Compute Engine bulkInsert API, and how Dynamic Workload Scheduler (DWS) handles provisioning for massive-scale batch jobs.

​You'll learn how to:

• ​Choose between NVIDIA GPU instance configurations for high-availability inference and clustered, rail-aligned infrastructure for distributed training

• ​Build automated fallback and dynamic instance routing so workloads keep running

• ​Improve unit economics by pairing Flex Start VMs and Spot instances with Flexible Committed Use Discounts (Flex CUDs)

• ​Use Capacity Planner to forecast compute demand and orchestrate resource allocation

​The presentation ends with live Q&A.

​Speaker: Chelsie Czop, Sr. Product Manager, AI Infrastructure, Google Cloud

​Who it's for: ML engineers, infrastructure and platform engineers, and technical leaders running AI training or inference at scale.

​When: Thu Oct 29, 10:00 AM PT