Rent L40 by the hour

Specs and price

VRAM

48 GB

Price

$1.50/hr

Availability (Live)

64 available

Multi-GPU configs

1x, 2x, 4x, 8x

What you can run on it

Every instance launches from a one-click template, with no image building or driver setup:

Serve an LLM (vLLM)

Deploy a ready-made model card (gpt-oss, Qwen3.5 and 3.6, Qwen3 Coder) or paste any Hugging Face model id. The launch is auto-configured for the model and this GPU, and it serves an OpenAI-compatible endpoint with a chat page built right into the instance view.

Run models with Ollama

Pull any open model, including GGUF quantizations, and use it through the Ollama API. Good for quantized weights and the Ollama tooling you already know.

JupyterLab

A CUDA PyTorch notebook environment, open in the browser in minutes.

ComfyUI

The node-based image and video workflow UI, with the Manager panel for installing custom nodes and models.

Web terminal

A browser bash shell into the instance for your own Docker-based workloads and serving stacks.

How GPU Cloud works

1. Pick a configuration

Choose the GPU count and a runtime storage target from the live catalog.

2. Pick a template

JupyterLab, ComfyUI, a vLLM model server, or a browser web terminal, ready in minutes.

3. Connect

Open the workload in the browser or call it through the authenticated EmpirioLabs connect endpoint.

Common questions

How is it billed?

Per second at the listed hourly rate, only while the instance is running. The rate is locked in when you deploy, and stopping or destroying the instance stops the charge.

What can I run on it?

One-click templates cover JupyterLab notebooks, ComfyUI, vLLM model serving with any Hugging Face model id, and a browser web terminal for your own workloads.

Is it available right now?

The facts above show live availability from the deploy catalog. When this GPU is out of stock, other GPUs with similar VRAM are usually available on the GPU Cloud page.

Can I chat with the model I deploy?

Yes. A one-click model instance includes a chat page right on the instance view, and the served endpoint is OpenAI-compatible for use from your code. Ollama instances expose the standard Ollama API.

Can I manage instances through the API?

Yes. Deploy, stop, and destroy instances under /v1/gpu on api.empiriolabs.ai, and reach the running workload through the authenticated EmpirioLabs connect endpoint. The full reference is in the GPU Cloud docs.