Every GPU on this page was tested within the last hour.
Rent NVIDIA GPUs by the second. The bandwidth, disk speed and uptime on each listing come from our own hourly tests, not from what the host typed in.
No machines are listed right now.
Measured listings appear here as soon as a verified host goes live. List a GPU
You pay the host's price
There is no platform fee on rentals. The number on the listing is the number on your bill, billed per second while the container runs.
Stopping never costs you the machine
When you stop, your GPU stays reserved for an hour at no GPU charge and your disk is kept. If your balance runs out we stop your instances and keep the disks for 72 hours.
Verification you can read
Each machine passes a published checklist, then an hourly proof of its GPUs, disk and network. Hosts and renters see the same numbers and thresholds.
Read the checklistFrom search to shell
- Pick a GPU by measured performance per dollar, reliability and network.
- Choose a template such as PyTorch, Jupyter, vLLM or ComfyUI, or bring any Docker image.
- Connect over SSH, open Jupyter or your app's URL, or use the browser terminal. Hosts never need to open ports, so every listing is reachable.
rgpu search --gpu RTX_4090 --sort perf-per-dollar rgpu launch --gpu RTX_4090 -t pytorch --wait rgpu ssh inst_01J…