NVIDIA adjusts hbm and nvlink bandwidth for rubin GPU systems, boosts network bandwidth

NVIDIA has revised the HBM and NVLink bandwidth for its Rubin GPU systems, while increasing network bandwidth. The HBM4 memory bandwidth for the Vera Rubin superchip has been reduced from 44TB/s to 38.5TB/s, and NVLink bandwidth has dropped from 7.2TB/s to 6TB/s. However, the cross-system network bandwidth has increased from 0.8TB/s to 0.9TB/s. The Vera Rubin NVL4 variant saw a more significant improvement, with its cross-system network bandwidth rising from 0.4TB/s to 1TB/s. In the non-superchip Rubin GPU SMX module solutions, HBM memory bandwidth decreased from 176TB/s to 130TB/s, and NVLink Switch bandwidth dropped from 28.8TB/s to 24TB/s. The bidirectional network bandwidth increased from 1.6TB/s to 1.8TB/s. NVIDIA also adjusted the memory specifications for all Vera CPUs, changing them from single 1.5TB LPDDR5X to a maximum of 1.5TB LPDDR5X.