Choose the technically eligible fleet and hand its operands to finance — stated inputs
Demand 48,000 tokens/s; resident set 180 GB/node; site IT-input budget 1,000 kW. A: 160 GB, 1,000 tokens/s, 20 kW/node. B: 186 GB, 800 tokens/s, 16 kW/node. C: 256 GB, 700 tokens/s, 15 kW/node. B memory/package/system ready weeks 12/14/18; transit 2 weeks; required arrival week 20.modeled
| Value kind | modeled — Interpret this value according to its displayed kind, scope, source, and as-of date. |
|---|---|
| Scope | Hypothetical complete-node offers at equal quality/latency with shared IT power. Unsupported tuples and dates are explained in the opening callout. Chapters 7.6 and 2.1 supply fit and schedule methods; C delivery remains unqualified. |
| As of | 2026-09 |
| Source | NVIDIA CUDA best-practices measurement method — method reference; guide-authored hypothetical scenario September 8, 2026. No supplier quote or test measurement. |
| Derivation | Exact teaching assumptions chosen to expose the stated constraint and its reversal; not estimates of market prices or actual equipment. |
| Review | checking…review by 2027-03-08 · standard cadence |
| Recorded changes | last 2026-09-16 |
| Claim id | guide2-t29-7-11-inputs |
Where the guide uses it
← Full numbers register — every date-stamped figure in the guide, with revision history.