Clockwork.io

Clockwork.io develops software for the networks and fault handling of large GPU clusters used for AI training and inference. Its about page says the company began in 2018 as TickTock Networks, based on research incubated at Stanford University, was renamed Clockwork Systems in 2021 and debuted publicly as Clockwork.io in 2022. The products are grouped in two areas. AI fault tolerance includes LinkPass and TorchPass, which are meant to keep jobs running through GPU errors, link flaps and node failures, and TorchPass Snapshots, which protect training progress without changes to training code. AI fabric observability includes FleetLens, which helps locate slow or failing jobs down to a link, switch or network card. The company states that the software works across InfiniBand, RoCE and Ethernet, with any GPU vendor and in any deployment, from hyperscale to on-premises. Its home page announces a 31 million dollar funding round and names LinkedIn, Together AI and WhiteFiber as adopters, and quotes an analysis from SemiAnalysis. Solutions pages address AI builders and neoclouds, and the site offers webinars, white papers, case studies and a blog.