NVIDIA Adjusts Rubin GPU Specifications, Lowering HBM and NVLink Bandwidth, Boosting Network Bandwidth

on October 26, citing IT Home, a comparison between NVIDIA's October 2026 document and a June version saved by a third party reveals that NVIDIA has lowered the HBM memory bandwidth and NVLink bandwidth for the Rubin GPU system, while increasing network bandwidth.

Specifically, the HBM4 memory bandwidth of the Vera Rubin superchip was reduced from 44 TB/s to 38.5 TB/s, and NVLink bandwidth dropped from 7.2 TB/s to 6 TB/s, while horizontal scaling network bandwidth increased from 0.8 TB/s to 0.9 TB/s. HBM capacity in the HGX NVL8 system was decreased from 2.3 TB to 2 TB, bandwidth from 176 TB/s to 130 TB/s, and NVLink Switch bandwidth also fell from 28.8 TB/s to 24 TB/s.

The report notes that lowering the HBM4 memory bandwidth threshold is expected to allow more memory products to meet the Rubin GPU specifications, thereby increasing final production volume. Additionally, NVIDIA has adjusted the memory specifications for the Vera CPU to support up to 1.5 TB of LPDDR5X per chip.