Run any AI model on a real GPU
16 GB to 80 GB VRAM available. ComfyUI, Automatic1111, LM Studio and more — on a Linux GPU machine with CUDA that you reach over SSH.
Supported AI tools
ComfyUI
Node-based Stable Diffusion workflows with full GPU acceleration
Automatic1111
A widely used Stable Diffusion web interface
LM Studio
Run large language models locally — Llama, Mistral, Mixtral
Ollama
CLI-based LLM runner. Pull and run models in seconds.
Fooocus
Simplified Stable Diffusion — great for beginners
KoboldCpp
Run GGUF models with GPU offloading
InvokeAI
Professional Stable Diffusion toolkit
LoRA training
Fine-tune Stable Diffusion models with your own data
VRAM requirements by model
| Model | VRAM needed | Starter (16 GB) | Standard (24 GB) | Pro (48 GB) | Power (80 GB) |
|---|---|---|---|---|---|
| Stable Diffusion XL | 8–12 GB | Yes | Yes | Yes | Yes |
| Flux.1 (dev/schnell) | 12–24 GB | Tight | Yes | Yes | Yes |
| Llama 3 8B (Q4) | ~6 GB | Yes | Yes | Yes | Yes |
| Llama 3 70B (Q4) | ~40 GB | No | No | Yes | Yes |
| Mixtral 8x7B (Q4) | ~26 GB | No | No | Yes | Yes |
| SDXL LoRA training | 16–24 GB | Tight | Yes | Yes | Yes |
Why TurboGPU for AI?
Browser studio or terminal
Every instance ships a browser-based Jupyter studio and full SSH access. Launch ComfyUI, LM Studio, or a notebook — whichever you prefer.
CUDA ready out of the box
Machines are Linux (Ubuntu) containers with the CUDA runtime, cuDNN and PyTorch pre-installed. PyTorch just works — no setup.
Use your existing workflow
The tools you already run — pip, conda, git — work as they do on any Linux GPU box. (Machines are containers, so Docker itself can't run inside them.)