News
NVIDIA increases memory bandwidth specifications for VR200 NVL72 AI systems
Tuesday, January 20, 2026 at 09:46 AM
NVIDIA has reportedly updated the specifications for its VR200 NVL72 systems, increasing memory bandwidth targets by 10%. This adjustment aims to improve performance metrics for AI infrastructure deployments.
Context
At CES 2026, NVIDIA revised the hardware specifications for its upcoming Vera Rubin VR200 NVL72 AI systems, increasing the total memory bandwidth to a massive 22 TB/s. This adjustment represents a significant leap from the initial 13 TB/s target set in early 2025 and a subsequent revision to 20.5 TB/s last September. By integrating faster HBM4 memory and refined interconnects, the company has effectively eliminated a key performance bottleneck for large-scale AI workloads.
This tactical upgrade allows NVIDIA to overtake AMD's upcoming Instinct MI400 series, specifically the MI455X, which offers 19.6 TB/s. By securing a 10% bandwidth advantage, NVIDIA aims to maintain its dominance in high-performance inference and training as hyperscalers transition to trillion-parameter models. The Vera Rubin platform is scheduled for initial shipments in the second half of 2026, promising up to a tenfold reduction in cost per token compared to current architectures.
Related Companies
Nvidia
NVDA