GPUs are steadily getting faster, but as enterprises, neoclouds and the entrenched hyperscalers look to get more efficiency out of them, the bottleneck is often the network — to the point where NVIDIA, for example, is now investing in silicon photonics to improve networking speed and resiliency. One key to improving existing networks is improved visibility into the state of these GPU clusters and the networks that connect the different systems.
Clockwork started out as a tool and service for synchronizing clocks across compute clusters. But…








