Nvidia cuts Vera Rubin bandwidth specs on product pages without notice
Per-GPU HBM4 bandwidth on the Vera Rubin NVL72 listing falls from 22 TB/s to 19.2 TB/s, with no press release issued to explain the change.
Nvidia has lowered the published memory and interconnect bandwidth figures for its Vera Rubin GPUs and HGX Rubin NVL8 systems, with no press release or public explanation accompanying the change, according to TechRadar. The edits appear on the company's own product pages.
On the Vera Rubin NVL72 page, each Rubin GPU is now listed at 19.2 TB/s of HBM4 bandwidth, down from 22 TB/s, and NVLink 6 bandwidth per GPU is listed at 3 TB/s, down from 3.6 TB/s. Per-GPU memory capacity stays at 288 GB, and the 50 PFLOPS of sparse NVFP4 inference is unchanged. At rack scale the aggregate memory bandwidth falls from 1,580 TB/s to 1,400 TB/s and the NVLink total from 260 TB/s to 216 TB/s. The two-GPU Vera Rubin Superchip figure drops from 44 TB/s to 38.5 TB/s.
The HGX Rubin NVL8 listing now shows 2 TB of HBM4 at 130 TB/s, down from 2.3 TB and 176 TB/s. NVLink Switch bandwidth for that system falls from 28.8 TB/s to 24 TB/s, and dense training figures slip from 280 to 270 PFLOPS of NVFP4 and from 140 to 130 PFLOPS of FP8. Scale-out networking is the only specification that rises, by 12.5% in most configurations.
The pages are now internally inconsistent. The per-GPU table still lists 288 GB at 22 TB/s with 3.6 TB/s of NVLink, figures that no longer match the printed system totals, and Nvidia's July deep dive into the Rubin architecture still promises 22 TB/s.
Earlier bandwidth targets
In April 2025, JEDEC published the HBM4 standard specifying up to 8 Gb/s per pin across a 2,048-bit interface, or 2 TB/s per stack. In March, a TechPowerUp report relayed a SemiAnalysis note saying SK hynix and Samsung were struggling to reach Rubin's final bandwidth figure, with early shipments expected nearer 20 TB/s. Nvidia has said Vera Rubin racks were already running at partners including CoreWeave, Google Cloud and Microsoft Azure, and the platform is positioned against AMD's MI455X and Helios.
Quick answers
What changed in Nvidia's Vera Rubin specifications?
Per-GPU HBM4 bandwidth on the Vera Rubin NVL72 page fell from 22 TB/s to 19.2 TB/s and NVLink 6 bandwidth per GPU from 3.6 TB/s to 3 TB/s. HGX Rubin NVL8 fell from 2.3 TB at 176 TB/s to 2 TB at 130 TB/s.
Did Nvidia announce the change?
No. The product pages were edited without a press release or public explanation, according to TechRadar.
Is per-GPU memory capacity affected?
No. Each Rubin GPU is still listed at 288 GB, and the 50 PFLOPS of sparse NVFP4 inference is unchanged.