per 32-node / 256-GPU NVIDIA scalable unit, read / write — DGX B300 SuperPOD 'Enhanced' storage tier
250 / 124 GB/sobserved
| 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 | NVIDIA B300 SU: 32 nodes, 256 GPUs; distinct from the 576-GPU GB200 NVL72 rack-scale planning SU. The Enhanced tier is 250 / 124 GB/s per SU and 2,000 / 992 GB/s at eight SUs; the Standard tier is 80 / 40 and 640 / 320. NVIDIA states these assume a mix of workloads and that requirements must be characterized per workload — they are reference tiers, not a per-GPU floor. |
| As of | 2025-11 |
| Source | NVIDIA DGX SuperPOD B300 Reference Architecture (Storage Architecture, Table 5) · Storage Architecture, Tables 4–5 (Standard and Enhanced performance requirements) |
| Derivation | 250 GB/s ÷ 256 GPUs ≈ 0.98 GB/s/GPU (Enhanced); 80 ÷ 256 = 0.3125 GB/s/GPU (Standard). Dividing by the separate 576-GPU SU gives 0.434 GB/s/GPU and mixes definitions. |
| Review | checking…review by 2026-11-19 · standard cadence |
| Recorded changes | last 2026-09-04 · 3 revisions tracked |
| Claim id | per-scalable-unit-read-write-dgx-b300-superpod |
Where the guide uses it
- 9.1 Storage in the AI Lifecycle: Why It Determines GPU Efficiency
- 9.2 Parallel & Distributed File Systems
- 9.3 NVMe Tiers, GPUDirect Storage & the CPU-Bypass Data Path
- 9.8 Sizing, Data Gravity & Resilience
← Full numbers register — every date-stamped figure in the guide, with revision history.