Vera Rubin per-GPU NVFP4: 50 PFLOPS inference (NVIDIA sparse specification), 35 PFLOPS training (dense specification)
~50 PFLOPS sparse (35 dense)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. |
|---|---|
| Scope | Per GPU. Per rack: 3,600 PFLOPS sparse inference and 2,520 PFLOPS dense training. |
| Caveat | Rubin Ultra's ~100 PFLOPS NVFP4 is a GTC 2025 roadmap figure. NVIDIA targets H2 2027; SemiAnalysis reported 2028 in July 2026. |
| As of | 2026-10 |
| Source | NVIDIA Vera Rubin NVL72 product page · NVFP4 inference and training rows, footnotes 2 (sparse) and 3 (dense) |
| Review | checking…review by 2027-01-16 · standard cadence |
| Recorded changes | last 2026-10-15 · 3 revisions tracked |
| Claim id | vera-rubin-per-gpu-fp4-inference-rubin-ultra |
Where the guide uses it
← Full numbers register — every date-stamped figure in the guide, with revision history.