Managed GPU jobs
In buildSubmit an image or a recipe with your inputs, budget and deadline. vgpu finds suitable capacity, runs the work, checkpoints it, recovers it if a machine is reclaimed, and returns the result.
Early access
GPU capacity that doesn't sit idle.
vgpu places your GPU work on the right card at the right price, across our own fleet and partner clouds. Run managed jobs, rent dedicated GPUs or serve models, with interruptible pricing, checkpoint-aware recovery and storage that survives preemption.
vgpu is in private build. Write to hello@vgpu.com and a person will reply.
What you can run
GPU leases
In buildRent a dedicated GPU container or VM with SSH access and persistent storage. Every offer states its runtime, memory per GPU and interruption terms before you start.
Inference endpoints
PlannedServe models behind a stable URL with warm capacity and autoscaling, so you pay for traffic instead of idle replicas.
Why vgpu
Interruptible capacity with real notice
Spare capacity at lower prices, plus a preemption notice your code can act on: a file, a local endpoint, a signal or a webhook. Adapters for existing AWS, GCP, Kubernetes and Slurm handlers are on the roadmap.
Storage that outlives the machine
Datasets and checkpoints live outside the GPU. When capacity is reclaimed, the replacement resumes from the last committed checkpoint instead of starting over.
Placement by memory per GPU
Jobs are matched to GPUs by the memory each card actually has, using measurements from real runs. Fewer out-of-memory crashes, less VRAM paid for and left empty.
Budgets you control
Every job carries a quote and a spend cap. vgpu never moves your work to a more expensive GPU than you approved.
One API across providers
Our own GPUs and partner clouds sit behind one API and SDK, so the same job runs wherever capacity is available.
Built on open source
SkyPilot, Kubernetes, Kueue and Temporal under the hood. Your workloads stay portable.
Built for
- Video and image generation that needs more GPU memory than one consumer card
- Fine-tuning and training jobs that can checkpoint and resume
- Batch inference and evaluation runs that can wait for cheaper capacity
- Teams with idle GPUs who want to put them to work
Want early access?
Tell us what you run, which GPUs you use today and what it costs. We'll show you where vgpu fits.
Email hello@vgpu.comHave idle GPUs?
Data centres and owners with spare capacity can list it with vgpu and keep their hardware busy.
Email partners@vgpu.com