Open WebUI + Ollama on a rented GPU
Your own private ChatGPT on a cloud GPU
Pick a GPU and it starts with Open WebUI and Ollama installed and running, behind its own private link. Pull Llama, Qwen, Gemma, DeepSeek or Mistral from inside it and chat. Nothing goes to an outside API; it all ends with the rental.
- H200 NVL 140 GBfrom $3.50/h
- RTX A6000 48 GBfrom $0.40/h
- RTX 5090 32 GBfrom $0.63/h
- RTX 4090 24 GBfrom $0.45/h
- RTX 4070 Ti SUPER 16 GBfrom $0.25/h
Measured, not quoted
On a rented RTX 4090
Measured on an RTX 4090 rented on ParalonCloud. On a 24 GB card a 32B model needs a short context to stay on the GPU: 6.3 tokens/s at Ollama's default, 40.8 with 4,096 tokens.
What you get
Installed, running, yours
Open WebUI and Ollama, ready
A tested Open WebUI version over a pinned Ollama, started for you; no login, the link is the key.
Any model from the Ollama library
Pull Llama, Qwen, Gemma, DeepSeek, Mistral or gpt-oss from the model selector, onto the machine.
Ollama's API over SSH
localhost:11434 and the OpenAI-compatible /v1 through your own tunnel, for scripts and editors.
Private by construction
Telemetry off, no outside API: chats and models stay on the machine and are deleted with the rental.
Which GPU
What fits on which card
| Model | 8–12 GB | 16 GB | 24 GB (RTX 4090) | 32 GB (RTX 5090) |
|---|---|---|---|---|
| llama3.2:3b, qwen3:4b, gemma3:4b | Fits | Fits | Fits (182–198 tok/s) | Fits |
| qwen2.5:7b, llama3.1:8b | Tight to fits | Fits | Fits | Fits |
| gemma3:12b, qwen3:14b | No | Fits | Fits | Fits |
| gpt-oss:20b, mistral-small3.2:24b | No | Tight | Fits | Fits |
| gemma3:27b | No | No | Fits (45.8 tok/s) | Fits |
| qwen3:32b, deepseek-r1:32b | No | No | Short context only | Fits |
From the models' download sizes, except where a cell gives a measured speed. More in Open WebUI and Ollama on a cloud GPU.
How it works
Three clicks to Open WebUI
- 01
Pick a GPU for your model
The list filtered for Open WebUI shows machines with 8 GB of VRAM or more; the table above says which model fits which card.
- 02
Rent with Open WebUI
The machine answers in about a minute; Ollama and Open WebUI install on top, a few minutes on a fast line.
- 03
Pull a model and chat
Type a model name in the selector, pull it from Ollama, and write. The first message loads it into VRAM.
FAQ
Open WebUI in the cloud
How do I run Ollama in the cloud?
Rent a GPU and choose Open WebUI: the machine starts with Ollama and Open WebUI running. Pull a model from inside Open WebUI, or with ollama pull in the terminal, and chat.
Which GPU do I need for Ollama?
The model's download size is roughly the VRAM its weights take, plus room for the conversation. 4B to 8B models run on 8 to 12 GB; 27B needs a 24 GB card like the RTX 4090; 32B models want 32 GB, or a short context on 24 GB.
Why does my model answer slowly?
Two common causes: a thinking model (qwen3, deepseek-r1) writing a long reasoning pass first, or a model too big for the card spilling onto the CPU. ollama ps shows 100% GPU when it fits.
Is it private?
The models run on the GPU you rented; nothing is sent to an outside API, telemetry is off, and everything is deleted when the rental ends. The link is the key: share it only with whom you would share the machine.
Can I use Ollama's API from my own computer?
Yes, on an SSH rental: add -L 11434:localhost:11434 to the SSH command and point any OpenAI client at http://localhost:11434/v1.
Can I start it from code or an AI agent?
Yes: "template": "openwebui" on POST /api/v1/rentals, or ask an agent with the ParalonCloud MCP server.
Open WebUI, ready in minutes
From $0.25 an hour, billed per minute. Stop whenever you like.