About vgpu

GPUs are expensive, and a lot of their time is wasted: machines wait idle between jobs, memory sits half-used, and a reclaimed spot instance often means starting a run over. vgpu exists to turn more paid GPU time into useful work.

We are building one service for running GPU workloads across our own fleet and partner clouds: managed jobs that checkpoint and recover on their own, dedicated GPU leases with persistent storage, and interruptible capacity that comes with a notice your code can act on.

vgpu is built by HYGO Inc., the team behind HYGO Shop and AIEH. We run large video, audio and model workloads ourselves, and vgpu is the platform we wanted for them.

Status

vgpu is in private build. This site describes what we are building; the API and console are not open yet. If you would like early access, tell us about your workloads.

Contact