Cluster track · Presentation
Overcoming AI's Power and Space Constraints with Scale-Across Deployments
- When
- Thursday, February 4, 20272:55 PM to 3:15 PM PT
- Where
- Room BSan Jose McEnery Convention Center
- Track
- ClusterArchitecting 1M GPU Clusters. Programmed by CloserStill with OCP
- Format
- Presentation
About this session
Every AI builder will eventually hit a wall. It isn't made of silicon. It is made of megawatts, floor space, and cooling. You can order more GPUs, but you cannot order a new power grid.
There is now a way around that wall, and it is running in production.
In this session, WhiteFiber and DriveNets share the build story of Project Redwood, the industry's first commercial, long-distance, scale-across deployment. Two NVIDIA GPU clusters, located 52 miles apart, now operate as a single logical supercluster. Validated at 111.2 Tbps across sites with 0.9 milliseconds of guaranteed latency, the network adds virtually zero overhead beyond the physical speed of light in fiber.
This is a case study from a live build, not a sales pitch. We will showcase real benchmarks comparing cross-site performance to same-site performance. We will also explain the engineering that makes 52 miles feel like 52 feet, from deep-buffered interconnect switches to granular load balancing across WAN links. Finally, WhiteFiber will share how scale-across networking changes capacity planning and what the next steps are for the industry.
Three takeaways for attendees:
Discover how to combine geographically separated sites into a single logical GPU supercluster to bypass local power and space constraints.
Learn how the physics is manageable when the network is built for it. Deep buffers, tuned congestion control, and granular load balancing can deliver lossless connectivity over long distances.
Review real-world benchmarking data proving cross-site AI training throughput can match same-site performance.
General rate, until Feb 2, 2027
Register for Digital Infrastructure World 2027
One pass covers this session and every other session, plus OCP Tech Day, the keynotes and the exhibit hall, on all three days.
- All Access$799
- Open Compute Project members$399
- Teams of 3 or more, each$599.25
More in the Cluster track
- From Hardware to Useful Capacity: Provisioning AI Infrastructure at Scale
- Beyond Physical Digital Twins: Why Logical Digital Twins Are Essential for AI Factory Design, Validation and Operations
- Node-to-Rack Efficiency for Real-World AI Inference: Arm and FuriosaAI
- One Rack, More Than One Accelerator: Lessons from Building an Open AI Rack