NVIDIA scale-encoding toy illustration: MSE 0.72 for E8M0 and 0.08 for E4M3
About 89% lower scale-encoding MSEderived
| Value kind | derived — Derived values are guide calculations; their result is only as fixed as the stated inputs, scope, and method. |
|---|---|
| Scope | One scale-encoding illustration. Does not establish an average model-tensor error reduction or a format-wide accuracy ranking. |
| As of | 2025-06 |
| Source | NVIDIA — Figure 4: scale-encoding illustration, not model-tensor per-block error. · Figure 4: scale-encoding illustration, not model-tensor per-block error |
| Derivation | (0.72 − 0.08) / 0.72 × 100 = 88.8889%. |
| Review | checking…review by 2027-06-24 · standard cadence |
| Recorded changes | last 2026-09-16 · 2 revisions tracked |
| Claim id | per-block-mse-reduction-from-nvfp4-s-e4m3-scale |
Where the guide uses it
← Full numbers register — every date-stamped figure in the guide, with revision history.