aiqbench analysis

CoreWeave, a cloud provider that deploys large GPU clusters, is publishing measured performance data for Nvidia's Vera Rubin NVL72 rack-scale system, which is useful because it reflects production-like deployment rather than only vendor specifications. Readers should check which workloads, precisions, and software stacks the numbers reflect, since those details determine whether they transfer to their own training or inference costs. Comparisons against the prior generation will matter more than absolute figures.