data throughput NVIDIA reports for xAI's Colossus on Spectrum-X Ethernet (adaptive routing + congestion control) — a vendor-reported deployment result
~95%observed
| Value kind | observed — Reported measurements, counts, and specifications keep the precision and scope stated by their source; an exact specification is not treated as a range. |
|---|---|
| Caveat | NVIDIA-reported Colossus deployment result under that workload and configuration, not a universal Ethernet, Spectrum-X, or scheduled/VOQ-fabric figure; Colossus is an adaptive-routing/telemetry RoCE fabric, not the scheduled-fabric category. Meta reports tuning RoCE and InfiniBand GenAI clusters to equivalent performance — no common-workload test crowns either transport. |
| As of | 2024-10 |
| Source | NVIDIA Newsroom, 'NVIDIA Ethernet Networking Accelerates World's Largest AI Supercomputer, Built by xAI' (Oct 2024) · Press release, Colossus network-performance paragraphs: 100,000 GPUs; zero flow-collision degradation/loss; 95% data throughput; 60% standard Ethernet |
| Review | checking…review by 2026-12-22 · standard cadence |
| Recorded changes | last 2026-08-24 · 3 revisions tracked |
| Claim id | effective-throughput-a-well-tuned-ai-ethernet |
Where the guide uses it
- 7.15 Deployment Velocity & Cabling at Scale
- 8.3 Network Silicon: Switch ASICs, NICs & DPUs
- 8.4 Scale-Out Fabric: Protocols, Standards & Transport
- 8.6 Congestion Control, Load Balancing & In-Network Compute
- 13.7 Network Fabric Commissioning & Validation
← Full numbers register — every date-stamped figure in the guide, with revision history.