UCPI-B200-LISTED Child Specification#

Status: proposed child specification, version 0.1.0-draft. Not launched. Prepared 14 September 2026. A child of the UCPI-LISTED-GPU family specification under the Urdais Compute Price Index family; everything not stated here is inherited from those documents unchanged. A different economic object from any accessible-offer child.

Instrument#

NVIDIA B200 as sold in HGX B200 systems (the SXM6 module), full physical device, per-accelerator allocation class. Sellers label device memory 180 GB or 192 GB for the same part, so memory does not gate identity for this instrument. GB200 and GB300 NVL rack-scale platforms, and the B300, are different products and are never admitted as B200 observations.

Admitted upstream SKUs#

Price of Compute canonical SKU B200. On first retrieval every seller row behind it carried a seller label of B200 SXM6, HGX B200 or B200 180 GB; no GB200 label was found. Admitted as Grade C identity, with that check recorded.

Seller topology evidence, 14 September 2026#

Verda: 1x B200 SXM6 180GB instance. Runpod: B200 180 GB listed per GPU with per-GPU host resources (Secure Cloud). Lambda: 1x B200 SXM6 instance exists, priced by quantity (1x, 2x, 4x, 8x), so the observation carries SELLER_PRICE_TIERED_BY_QUANTITY. Hyperstack: B200 listed per GPU; not present in the source's B200 rows on first retrieval. Massed Compute: B200 SXM6 sold only as an eight-GPU node, and the source's per-accelerator figure is that node price divided by eight; excluded WHOLE_NODE_REQUIRED. CoreWeave: HGX B200 node, excluded WHOLE_NODE_REQUIRED. Vast.ai: platform aggregate, excluded.

Version History#

0.1.0-draft, 14 September 2026: initial child. No production effective date.