Nvidia's Rubin ships 13% less memory bandwidth than the launch deck promised
Nvidia trimmed Vera Rubin's HBM4 bandwidth from 22 to 19.2 TB/s and NVLink 6 from 3.6 to 3.0 TB/s without an announcement. The eight-GPU NVL8 lost 26% of its memory bandwidth. Here is why bandwidth, not FLOPS, now sets AI infrastructure costs.