Nvidia forced to dial back key bandwidth targets on Vera Rubin hardware — tweaks will affect target specifications for memory and interconnect fabric
Reported by TechRadar and shattered.io2 sources ↓
HBM4 bandwidth dropped 13% while NVLink dropped 17%, and it took six weeks for anyone to notice.
Key points
- Vera Rubin NVL72 now lists 19.2 TB/s of HBM4 bandwidth and 3 TB/s of NVLink 6 per GPU
- NVL72 rack totals fell to 1,400 TB/s of memory bandwidth and 216 TB/s of NVLink bandwidth
- The HGX Rubin NVL8’s listed memory bandwidth fell 26%, to 130 TB/s, and capacity fell to 2 TB
- Nvidia gave no explanation for the revisions
Sources
Details
TechRadar reports that Nvidia changed its Vera Rubin product-page specifications without an announcement. For the NVL72, per-GPU HBM4 bandwidth is now listed at 19.2 TB/s, down from 22 TB/s, and NVLink 6 at 3 TB/s, down from 3.6 TB/s. The rack’s aggregate figures fell to 1,400 TB/s and 216 TB/s, respectively.
The HGX Rubin NVL8 saw larger changes: its listed HBM4 capacity fell from 2.3 TB to 2 TB, and its system memory bandwidth from 176 TB/s to 130 TB/s. Its NVLink Switch bandwidth decreased from 28.8 TB/s to 24 TB/s. Nvidia’s listed NVL8 training figures also declined, while scale-out networking figures rose in most configurations.
TechRadar says comparisons of archived product pages place the changes between September 1 and September 11, and that the lower figures had been published for at least six weeks before they were reported. Some Nvidia materials still show the former per-GPU figures, which do not match the revised system totals. Nvidia has not explained the revisions or said how the hardware achieves the changed specifications.
Written from coverage as of
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh