NVIDIA Revises Vera Rubin HBM4 and NVLink Specifications

NVIDIA has revised published specifications for its next-generation Vera Rubin AI computing platform, lowering HBM4 memory bandwidth and sixth-generation NVLink bandwidth in multiple configurations while increasing scale-out networking capabilities. The changes were not accompanied by a public announcement, according to the source.

The revisions were identified by comparing NVIDIA’s current Vera Rubin NVL72 specifications with archived versions of the same webpage. Wayback Machine snapshots dated September 1 and September 11, 2026 show that the specifications changed sometime between those dates.

The source reports that HBM4 bandwidth was reduced by up to 26%. It also says sixth-generation NVLink speeds were reduced, but does not provide the size of those reductions in the supplied material. The changes therefore appear to have been made more than a month before the report, rather than recently.

Guru3D and ITHome highlighted the revisions, according to the source. The supplied information does not specify the revised bandwidth figures for individual configurations.