reply Through extreme co-design, GB300 NVL72 uses an innovative copper backplane to expand the scale-up domain to 72 versus 8 on Hopper. This allows optimizations like wide expert parallelism, which allows each GPU to have only a subset of experts, resulting in: 🟠 Way lower weight-loading pressure 🟠 Way higher arithmetic intensity through the reuse of weights across many tokens (2/3)