NVIDIA DGX SuperPOD with DGX Vera Rubin NVL72: Enhanced storage tier per scalable unit of 8 systems (576 GPUs), aggregate read / write
280 / 140 GB/srecommendation
| Value kind | recommendation — Recommendations are design guidance, not measured facts; validate them against the project boundary conditions. |
|---|---|
| Scope | NVIDIA reference storage guideline, not a measured result or a per-GPU floor. The Standard tier is 90 / 45 GB/s per SU; at four SUs Enhanced is 1,120 / 560 GB/s and Standard 360 / 180. Enhanced read works out to about 0.49 GB/s per GPU, the same as NVIDIA's DGX B300 Ethernet reference architecture. Large vision datasets (over 30 TB) call for 4 GB/s per GPU read in all three NVIDIA reference architectures. |
| As of | 2026-09 |
| Source | NVIDIA DGX SuperPOD Reference Architecture featuring DGX Vera Rubin NVL72 (storage guidelines, Table 6) · Table 6: single-SU aggregate read 90 / 280 GB/s and write 45 / 140 GB/s (Standard / Enhanced) |
| Review | checking…review by 2027-03-15 · standard cadence |
| Recorded changes | last 2026-10-16 |
| Claim id | per-scalable-unit-read-write-dgx-vera-rubin-nvl72-superpod |
Where the guide uses it
← Full numbers register — every date-stamped figure in the guide, with revision history.