The world's top-performing system for graph processing at scale was built on a commercially available cluster. NVIDIA last month announced a record-breaking benchmark result of 410 trillion traversed edges per second (TEPS), ranking No. 1 on the 31st Graph500 breadth-first search (BFS) list. Performed on an accelerated computing cluster hosted in a CoreWeave data center in Dallas, the winning run used 8,192 NVIDIA H100 GPUs to process a graph with 2.2 trillion vertices and 35 trillion edges. This result is more than double the performance of comparable solutions on the list, including those hosted in national labs.

To put this performance in perspective, say every person on Earth has 150 friends. This would represent 1.2 trillion edges in a graph of social relationships. The level of performance recently achieved by NVIDIA and CoreWeave enables searching through every friend relationship on Earth in just about three milliseconds. Speed at that scale is half the story—the real breakthrough is efficiency. A comparable entry in the top 10 runs of the Graph500 list used about 9,000 nodes, while the winning run from NVIDIA used just over 1,000 nodes, delivering 3x better performance per dollar.