Clockwork.io has launched TorchPass Workload Fault Tolerance, a software solution that eliminates costly GPU training failures through Live GPU Migration technology. The system allows AI training workloads to continue running through hardware failures, network disruptions and node crashes without requiring checkpoint restarts. The company claims TorchPass can save over $6 million annually in a typical 2,048-GPU deployment by reducing wasted training progress by 95%. In large clusters, it cuts lost time from approximately three hours per day to under ten minutes. Independent testing by SemiAnalysis found TorchPass delivered faster fault-tolerant performance than standard checkpoint-restart approaches and higher Model FLOPs Utilisation than leading open-source alternatives. The solution typically completes recovery in approximately three minutes whilst training continues uninterrupted. TorchPass is now available as part of Clockwork.io's FleetIQ platform.

Clockwork Raises $20.57 Million in Funding