Nvidia has been iterating on its upcoming Vera Rubin superchip as it prepares for first server shipments in late summer, with the company significantly increasing the memory bandwidth of the VR200 NVL72 system over the past year to better match AMD’s Instinct MI400 accelerator family. According to SemiAnalysis, Nvidia’s initial target for the ‘Vera Rubin’ VR200 NVL72 system was 13 TB/s in March, which was upgraded to 20.5 TB/s by September. At CES 2026, Nvidia confirmed that the VR200 NVL72 system is now operating at 22 TB/s of bandwidth, bringing it ahead of AMD’s Instinct MI455X accelerator, which has 19.6 TB/s, after originally trailing in system bandwidth. Nvidia is reported to have closed this gap by adopting faster DRAM and refining interconnects that link CPUs, GPUs, and the broader system.
AMD has been positioning its Instinct MI400 lineup directly against Nvidia’s Vera Rubin, emphasizing advantages in memory capacity and scale-out connectivity while targeting comparable compute throughput and bandwidth. In November, AMD compared the MI400 lineup to Nvidia’s upcoming ‘Vera Rubin’ series, claiming similar compute performance and memory bandwidth but 1.5 times higher memory capacity and scale-out bandwidth. The company is framing MI400 as a generational leap over its current MI350 accelerators, with a particular focus on large scale Artificial Intelligence workloads and dense high performance computing.
On the raw performance side, AMD claims it will deliver up to 40 FP4 and 20 FP8 PFLOPs, roughly twice the compute performance of the current MI350. The GPUs also transition to HBM4 memory from HBM3e, increasing capacity from 288 GB to 432 GB and raising total bandwidth from 8 TB/s to 19.6 TB/s. Each GPU provides 300 GB/s of scale-out bandwidth and adds broader Artificial Intelligence data format support along with expanded Artificial Intelligence pipelines, underscoring AMD’s focus on training and inference efficiency. AMD is planning two key models in the series: the Instinct MI455X, aimed at large-scale Artificial Intelligence training and inference deployments, and the MI430X, targeted at high performance computing and Artificial Intelligence, setting up a direct competitive clash with Nvidia’s Vera Rubin platforms as both enter the market.
