GPU Marketplace
Rent GPU compute by the hour — RTX 4090, A100 (40GB/80GB) and H100 — with instant provisioning and no long-term commitment. GPUs are listed by node operators and rented directly through the console (programmatic access goes over the signed gRPC API, nexlayer.gpu.v1.GpuService); billing is metered hourly against your account balance.
Renting a GPU
- Open GPU → Marketplace to browse available listings, filtered by GPU model, region and price.
- Pick a listing and choose a template (see below) or start with a bare Ubuntu image.
- Click Rent — the instance provisions in the background; status moves from
provisioningtoactive. - Once active, connect over SSH or the built-in web console, same as a regular server.
Marketplace
Available GPUs are listed in the console under GPU — browse offers and prices there before renting.
Templates
Rentals can start from a pre-built template instead of a bare image — the container image, ports and environment are already configured:
| Template | Use case |
|---|---|
| PyTorch | General deep learning, custom training scripts |
| TensorFlow | TensorFlow/Keras workloads |
| vLLM | High-throughput LLM inference serving |
| ComfyUI | Node-based Stable Diffusion workflows |
| SD WebUI (Automatic1111) | Stable Diffusion image generation |
| Jupyter | Interactive notebooks |
Each template declares a minimum VRAM requirement, so the console only offers templates that fit the GPU you're renting.
Billing
Rentals are billed hourly, metered against your prepaid balance while the rental is active. Stopping or deleting a rental stops the meter. Metrics (GPU utilization, memory, temperature, power draw) are recorded and viewable per rental for the duration you rented it.
Networking
A GPU rental can be attached to a private network like any other server, so it can reach your databases or other private resources over the tenant overlay — see Private Networking.