Report
SF Autoresearch adds InfiniBand clusters in internal preview
SF Compute says each cluster node has eight H100 GPUs and bills by the minute once ready.
TLDR
SF Compute has introduced InfiniBand clusters in Autoresearch for distributed training across H100 machines on a private network. It says each node has eight GPUs and bills at its ordinary rate once ready, with no minimum charge. The clusters remain in internal preview, and customers can register interest in access.
Combined views
15
1 Source, first seen ago