Nvidia and Coreweave set Graph500 record with H100 gpu cluster

Nvidia and Coreweave achieved a record-breaking 410 trillion traversed edges per second on the 31st Graph500 breadth-first search benchmark using a commercially available cluster of 8,192 H100 gpus hosted in Dallas. The result showcases a gpu-only, full-stack approach to large-scale graph processing that doubles the performance of comparable systems while using far fewer nodes and lower cost.