Rent RTX 4090 16 GB

Specifications

Video memory

16 GB

Memory type

GDDR6X

Architecture

Ada Lovelace

Class

Consumer

What it is good for

Estimates for short context and typical settings. See the guide on memory for LLMs. How much VRAM an LLM needs: a table and a rule of thumb

Where it is available now

How to start

1. Pick a server with the memory you need and a price you like in the catalog. Memory and type filters are on the catalog page.
2. Launch it with a template: Ollama for chatting with LLMs, ComfyUI or Forge for images and video, PyTorch + Jupyter for experiments.
3. Work in the browser or over SSH. Stop the server when you are done: GPU minutes stop being billed.

Pricing

Ready AI apps for the RTX 4090 16 GB

• Open WebUI + OllamaYour own ChatGPT with any model from Ollama or Hugging Face: name it at launch or pull it in the chat.
• ComfyUIImage and video generation: FLUX, SDXL, Wan, HunyuanVideo. Nodes and ready workflows.
• Stable Diffusion WebUI ForgeThe familiar A1111 interface with Forge acceleration, ControlNet and extensions.
• GPT-OSS 20B UncensoredOpenAI's open model with refusals removed. Fast, fits in 16 GB of VRAM.
• Wan2GPVideo generation with Wan 2.2, Hunyuan and LTX in one interface, light on VRAM.
• FluxGymFLUX LoRA training in a few clicks: upload photos and captions, get a model.
• Kohya SSTrain LoRA and DreamBooth models for SDXL and FLUX.
• Unsloth StudioFine-tune LLMs (Qwen, Llama, Gemma) in the browser: dataset, training, testing, export.

Run it on gpu.nz

• Stable Diffusion and FLUX on a rented GPU: Forge and SDXL online
• ComfyUI online on your own server: FLUX, Wan and HunyuanVideo
• JupyterLab with a GPU for rent: PyTorch, CUDA and data science
• GPU rendering on a rented server: Blender Cycles and 3D by the minute
• Video generation on a rented GPU: Wan and HunyuanVideo in ComfyUI

Questions

How much does a RTX 4090 16 GB rental cost?

From about $0.766 per GPU-hour at the lowest current offer. Prices change with availability, so the live figure above is the one to trust.

How is billing calculated?

Per minute while the server runs, from your balance. The price is fixed at launch. A server that fails to start is not billed.

Do I pay for the GPU when the server is stopped?

No GPU time is billed while the server is stopped. The disk is still billed, so keep only the data you need on it.

Is my data kept?

Data on the disk stays while the server exists. Deleting the server deletes its disk too, so download results first.

Which template should I start with?

Ollama for chatting with language models, ComfyUI or Forge for images and video, PyTorch with Jupyter for training and experiments, Ubuntu with CUDA when you want to install everything yourself.

Related models

• RTX 5090 32 GB$0.828/h
• RTX 5080 16 GB$0.381/h
• RTX 5070 Ti 16 GB$0.335/h
• RTX 5070 12 GB$0.28/h
• RTX 5060 Ti 16 GB$0.231/h
• RTX 5060 Ti 8 GB$0.231/h