Lambda runs All-Reduce across 10,368 NVIDIA GB300 GPUs

TheZachMueller · x · 2026-08-22

Lambda announced the successful execution of an All-Reduce operation across a cluster of 10,368 NVIDIA GB300 GPUs. The setup consists of 9 pods and 144 racks. This demonstrates their engineering capability in managing and interconnecting hyperscale GPU clusters, typically used for large-scale model training.

Related event: Lambda Runs All-Reduce Test Across Over 10,000 GB300 GPUs(2 posts)→

Original post →

More from Infra

Infra channel →