NVIDIA's reference training back-end is a non-blocking, two-tier, eight-rail-optimized fat tree; inline L7 inspection costs goodput there
1:1 non-blockingobserved
| 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. |
|---|---|
| Scope | Reference design, not a universal rule. Deployments may oversubscribe above the pod (for example Meta's 1:7 above non-blocking 3,072-GPU pods; see Chapter 8.1). |
| As of | 2025 |
| Source | SemiAnalysis / NVIDIA · Optimizing the Back-end Network: NVIDIA reference topology is a two-tier, eight-rail fat tree with non-blocking connectivity |
| Review | checking…review by 2027-04-16 · standard cadence |
| Recorded changes | last 2026-06-29 |
| Claim id | training-back-end-fabric-design-sub-2-us |
Where the guide uses it
← Full numbers register — every date-stamped figure in the guide, with revision history.