GB200 NVL72 in-domain all-reduce busbw across 2–32 GiB buffers (one provider's 72-rank run); scale-out gate set separately
~806–928 GB/s (2–32 GiB; sweep average ~872 GB/s)observed
| 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 | One provider's validation run: nccl-tests all_reduce_perf, 72 ranks in one GB200 NVL72 domain (18 nodes × 4 GPUs), 2–32 GiB buffers, out-of-place. Busbw exceeds 900 GB/s only at 16 GiB and above. The source does not state the NVLS setting. The scale-out gate is derived separately from NICs, rails, topology, protocol efficiency, message size and collective model. |
| As of | 2026-04 |
| Source | Crusoe Cloud, How-To Setup GB200 NVL72 Rack on CMK Cluster and Run NCCL Performance Validation (updated 2026-04-30) |
| Review | checking…review by 2027-04-16 · fast cadence |
| Recorded changes | last 2026-10-16 · 3 revisions tracked |
| Claim id | in-domain-all-reduce-busbw-on-gb200-nvl72 |
Where the guide uses it
← Full numbers register — every date-stamped figure in the guide, with revision history.