Data that keeps up with reasoning
Episode details
Episode description
Modern reasoning models don't just need more compute, they need storage that can feed GPUs data as fast as they can process it. In this session, CoreWeave's Emma Ramos (Senior Product Marketing Manager) and Jeff Braunstein (Principal Product Manager, Storage & Data Services) break down how CoreWeave's storage portfolio is purpose-built to keep pace with AI training and inference at scale. You'll learn about: - CoreWeave's full storage portfolio: AI Object Storage, Distributed File Storage, Dedicated Storage, and GPU Local Storage - CoreWeave AI Object Storage: exascale, S3-compatible storage with read-after-write distributed caching and up to 7 GB/s per GPU throughput - LOTA (Local Object Transport Accelerator): CoreWeave's patented technology for low-latency data delivery directly to GPU nodes - How automated, usage-based billing lowers storage costs for inactive data without tiering, data movement, or archive access fees - Techniques for maximizing GPU utilization and goodput during training runs - How to monitor object storage operations and performance in production - Whether you're training large-scale models or running high-throughput inference, this session shows how the right storage architecture removes data bottlenecks so your GPUs stay fed and fully utilized. Learn more about CoreWeave's storage solutions: https://utm.io/utbVx
Watch also from the same channel
More episodes from this channel
You might be interested in
