Keep the lowest precision that clears the quality contract — stated inputs
10,000 evaluation items; BF16 correct 9,000; FP8 correct 8,940; NVFP4 correct 8,880; quality floor 89.0%; required 1,000 tokens/s; p99 inter-token ceiling 50 ms; FP8 1,200 tokens/s at 45 ms; NVFP4 1,600 tokens/s at 42 ms.modeled
| Value kind | modeled — Interpret this value according to its displayed kind, scope, source, and as-of date. |
|---|---|
| Scope | Exact synthetic evaluation, no statistical generalization. Unsupported outcomes and gate ranges are explained in the opening callout; Chapter 1.7 owns workload definition. |
| As of | 2026-09 |
| Source | NVIDIA, Introducing NVFP4 — method reference; guide-authored hypothetical scenario September 8, 2026. No supplier quote or test measurement. |
| Derivation | Exact teaching assumptions chosen to expose the stated constraint and its reversal; not estimates of market prices or actual equipment. |
| Review | checking…review by 2027-03-08 · standard cadence |
| Recorded changes | last 2026-09-16 |
| Claim id | guide2-t29-7-10-inputs |
Where the guide uses it
← Full numbers register — every date-stamped figure in the guide, with revision history.