What You Can Do
Three ways to use RunGPU — from one-click inference to full creative workspaces.
How It Works
Browse GPUs
Filter by VRAM, price, GPU model, and region.
Choose Your Model
HuggingFace URL, well-known model, or custom Docker image.
Agent Pulls & Runs
The host's agent downloads the model, runs it on their GPU.
Get Your Output
Video, image, or API response — uploaded to cloud storage.
Run Any AI Model
Paste a HuggingFace URL, pick a popular model, or bring your own Docker image. No setup, no config files — just GPU and go.
Any HuggingFace Model or Space
Zero ConfigPaste any HuggingFace Space URL and we handle the rest — resolve the Docker image, pull it, mount your GPU, and run it. Works with 400,000+ models on HuggingFace.
One-Click Models
— pre-built, optimized, ready to runLTX-Video
Text → video
Video GenWan2.1
Text → video
Video GenStable Diffusion
SDXL, SD 1.5
Image GenFLUX
FLUX.1 models
Image GenComfyUI
Node-based UI
WorkspaceLlama 3
Meta 8B/70B
LLMMistral
7B/Large
LLMWhisper
Speech → text
AudioJupyter
PyTorch notebook
WorkspaceCustom Docker
Your own image
AnyCustom LoRAs & Workflows
Upload LoRAs from Civitai, custom ComfyUI workflows, or fine-tuned checkpoints. They're automatically downloaded and mounted into your workspace.
Any Docker Image
Bring your own Docker image from any trusted registry. NVIDIA NGC, GitHub Container Registry, Docker Hub official images — all supported.
Smarter Than a Cloud Provider
Under the hood, RunGPU automatically optimizes cost, performance, and reliability — so you don't have to.
Smart Routing
Your job is automatically matched to the best GPU based on model requirements, VRAM, price, and cached models. Hosts with your model already loaded get priority — instant cold starts.
Auto-Failover
If a host goes offline mid-job, your workload automatically migrates to an equivalent GPU. Billing pauses during downtime. You only pay for actual compute.
Multi-Runtime Orchestration
The platform detects what your model needs — Ollama for LLMs, Docker for video/image gen, workspace mode for ComfyUI — and configures the right runtime automatically.
Cost Optimization
Real-time pricing across all available GPUs. The system finds the cheapest option that meets your requirements — or you can set a max budget per project.
Zero-Trust Security
Every container is sandboxed. No privilege escalation, no host filesystem access, fork bomb protection. Hosts control exactly which image sources are allowed.
Real-Time Alerts
Email and in-app notifications for every event — job started, completed, migrated, host offline. Full audit trail of every dollar spent and earned.
How a job flows through the platform
You
pick a model
Router
finds best GPU
Agent
pulls & runs
Output
delivered to you
Get Started in 3 Minutes
Whether you're renting or hosting — here's how to go from zero to running.
Rent a GPU
For AI developers & creators
Sign up & add credits
Create an account and purchase credits with your card. Pay as you go — no subscriptions.
Browse & pick a GPU
Filter by VRAM, price, model, or region. See real-time availability and ratings.
Choose your model
Select a pre-built model, paste a HuggingFace URL, or provide your own Docker image.
Launch & get results
Your job runs on the host's GPU. Get video, images, text, or a live workspace URL.
Host Your GPU
For GPU owners who want to earn
List your GPU
Select your GPU model, set your price per hour, and choose your region. Takes 30 seconds.
Install the agent
One-line install on macOS, Linux, or Windows. The agent handles Docker, billing, and security.
Connect Stripe
Set up Stripe Connect to receive payouts directly to your bank account. Tax forms handled.
Earn automatically
Your GPU earns 85% of every rental. Billing, fault tolerance, and security are all managed for you.
Or use the API directly
// 1. Deploy a model (any HuggingFace URL or Docker image)
const deploy = await fetch("https://www.rungpu.io/api/deploy", {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({
user_id: "your-user-id",
url: "stabilityai/stable-diffusion-xl-base-1.0", // or any HF model
})
});
const { deployment } = await deploy.json();
// → deployment.endpoint_url = "https://www.rungpu.io/api/gpu/abc-123"
// 2. Send requests to your endpoint
const result = await fetch(deployment.endpoint_url, {
method: "POST",
headers: { "Authorization": "Bearer tk_...", "Content-Type": "application/json" },
body: JSON.stringify({ prompt: "A cat in space" })
});Built for Trust
Container Sandboxing
Every workload runs in an isolated container with no-new-privileges, pids-limit, and blocked host filesystem access.
Trusted Sources Only
Docker images and custom files only from verified sources: HuggingFace, Civitai, GitHub, NVIDIA NGC.
Server-Side Billing
Billing uses server timestamps — hosts can't inflate time. Atomic transactions prevent race conditions.
Open-Source Agent
The GPU agent is open source. Hosts can audit every line of code running on their machine.
Have a GPU? Start Earning.
List your GPU on the marketplace. Install our lightweight agent. Earn 85% of every rental — we handle billing, fault tolerance, and security.
85%
Your earnings
30s
Setup time
24/7
Auto-managed
FAQ
What models can I run?
Any model with a Docker image — LTX-Video, Wan2.1, Stable Diffusion, FLUX, Llama, Mistral, ComfyUI, or your own custom image from HuggingFace.
How does billing work?
Per-hour billing with server-side timestamps. You only pay for actual compute time. If a host goes offline, billing stops instantly.
Is it safe to list my GPU?
Yes. Every container is sandboxed with no-new-privileges, pids-limit, and blocked host filesystem access. Only trusted Docker images are allowed by default.
Can I use custom LoRAs?
Yes. Upload LoRAs from Civitai or HuggingFace, ComfyUI workflows, and custom checkpoints. They're mounted into the container automatically.
What if the host goes offline?
Your rental auto-migrates to an equivalent GPU. If no migration target is available, the rental terminates and you're only charged for uptime.
Is the agent open source?
Yes. The GPU agent is open source so hosts can audit every line of code running on their machine.