HBM capacity per GPU package: H100 80 GB, B200 SXM 180 GB, B300 SXM 288 GB, Rubin 288 GB HBM4 (shipping); Rubin Ultra announced at ~1 TB HBM4E (roadmap)
H100 80GB → Rubin Ultra 1TBforecastcontested
| Value kind | forecast — Forecasts are time-bound outlooks, not specifications; test decisions across plausible scenarios. |
|---|---|
| Scope | Use 180GB for the current NVIDIA HGX B200 SXM product specification; an unqualified 192GB figure does not match NVIDIA's current usable-memory product table. Rubin Ultra announced at ~1 TB HBM4E per four-die package (GTC 2025); SemiAnalysis reported two compute dies (2026-06-30) and 192 GB per GPU (2026-09-13); NVIDIA has confirmed neither (Chapter 7.2). |
| Caveat | Shipping points (H100 80 GB, B200 180 GB, B300 288 GB, Rubin 288 GB HBM4) are observed; the Rubin Ultra ~1 TB HBM4E endpoint is NVIDIA's GTC 2025 announcement. In August 2026 The Information reported 192–256 GB test configurations and in September 2026 SemiAnalysis reported 192 GB per GPU; NVIDIA has confirmed neither. |
| As of | 2026-10 |
| Source | NVIDIA · Table 1: HGX H200, B200 and B300 GPU and per Node & Memory specifications |
| Review | checking…review by 2027-01-17 · standard cadence |
| Recorded changes | last 2026-10-19 · 5 revisions tracked |
| Claim id | per-gpu-hbm-capacity-trajectory-b200-192gb-b300 |
Where the guide uses it
Not quoted in a chapter yet; it is kept in the curated register.
← Full numbers register — every date-stamped figure in the guide, with revision history.