NVIDIA Quietly Cuts Vera Rubin HBM4 Bandwidth by Up to 26%
NVIDIA has silently revised the published specifications of its next-generation Vera Rubin AI platform. According to archived page comparisons, HBM4 memory bandwidth was reduced by up to 26% in some configurations and sixth-generation NVLink speeds were lowered, while scale-out networking capabilities were increased. The changes appear to have been made between September 1 and 11, 2026.
