Whole GPUs, passed through
You get the physical card with full VRAM. Install your own drivers, CUDA version and container runtime.
No vGPU slicingWhole GPUs on single-tenant hardware — no slicing, no noisy neighbours and no meter running by the hour. Train, fine-tune, serve or render on a machine nobody else touches.
Plans
All unmetered. Multi-GPU and custom chassis are built to order.
Unlimited traffic
Unlimited traffic
Unlimited traffic
Unlimited traffic
What you get
You get the physical card with full VRAM. Install your own drivers, CUDA version and container runtime.
No vGPU slicingNVMe Gen4 and Gen5 scratch space for datasets and checkpoints, with 10–25 Gbps ports to move them in and out.
NVMe Gen4 / Gen5High-density racks with A+B feeds sized for sustained full-load draw, not burst.
Sustained full loadOperating systems
Whatever you install. You have root and the physical GPU, so PyTorch, TensorFlow, JAX, vLLM, ComfyUI and anything else that runs on NVIDIA hardware will work. We can pre-install the NVIDIA driver and container toolkit on request.
In-stock models are handed over the same day. Multi-GPU servers are built to order and take around ten days including burn-in under full load.
Bitcoin, Monero, Ethereum, USDT and Litecoin, plus cards and bank transfer in EUR. Crypto payments need no personal details. Card and bank payments carry the name your bank attaches to them — we cannot remove that.
From €79 a month. No setup fee, no ID, no lock-in on monthly terms.