per NVIDIA DGX B300 SuperPOD scalable unit (64 nodes / 512 GPUs in the Spectrum-X Ethernet RA; 72 nodes / 576 GPUs in the XDR InfiniBand RA), read / write: 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 publishes two DGX B300 SuperPOD reference architectures with the same per-unit storage table but different scalable units: 64 DGX B300 systems and 512 GPUs in the Spectrum-X Ethernet, DC-busbar RA, and 72 systems and 576 GPUs in the XDR InfiniBand RA (Architecture page, Table 3). Match the unit to the RA you are buying against. The Enhanced tier is 250 / 124 GB/s per unit and 2,000 / 992 GB/s at eight units; the Standard tier is 80 / 40 and 640 / 320. NVIDIA says the tiers assume a mix of workloads and that requirements must be characterized per workload, so they are reference tiers, not a per-GPU floor. |
| As of | 2026-09 |
| Source | NVIDIA DGX SuperPOD B300 Reference Architectures: Spectrum-4 Ethernet and DC busbar power (Storage Architecture, Table 5; updated 2025-11-19) and XDR InfiniBand (same table; updated 2026-09-02) · Architecture page Table 3 (scalable units, both RAs); Storage Architecture Tables 4-5 (Standard and Enhanced performance requirements). XDR RA: https://docs.nvidia.com/dgx-superpod/reference-architecture/scalable-infrastructure-b300-xdr/latest/storage-architecture.html |
| Derivation | Enhanced read: 250 / 512 = 0.49 GB/s/GPU (Ethernet RA); 250 / 576 = 0.43 (XDR RA). Standard read: 80 / 512 = 0.16; 80 / 576 = 0.14. The eight-unit figures give the same ratios. |
| Review | checking…review by 2027-03-02 · standard cadence |
| Recorded changes | last 2026-10-16 · 4 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.