We heard someone dropped about 10k NVIDIA GB300's somewhere? Don't worry, we made good use of it. This test marks the largest cluster-wide All-Reduce on Lambd

We heard someone dropped about 10k NVIDIA GB300's somewhere? Don't worry, we made good use of it. This test marks the largest cluster-wide All-Reduce on Lambda ever (2.5x the prior record). 9 pods, 144 racks, 2592 nodes, and 10,368 GPUs. [📷](https://pbs.twimg.com/media/HQRcW5AbcAAHoHx.jpg)