Dedicated GPU power for AI, billed by the hour
Servers with a dedicated NVIDIA Blackwell GPU (not shared) and AMD EPYC Milan CPU up to 3.65 GHz. Built for LLM inference, fine-tuning and image generation, with your data in the EU.
NVIDIA Blackwell GPU
Blackwell-generation RTX PRO cards, dedicated entirely to your server, with up to 48 GB total VRAM.
AMD EPYC Milan CPU
Dedicated AMD EPYC Milan processors up to 3.65 GHz, for fast pre-processing and data pipelines.
Dedicated resources
The GPU and CPU are yours alone, never shared, so performance stays constant and predictable.
Pay by the hour
Start the server for as long as you need and stop it when you are done. No subscription, no monthly commitment.
GPU plans for AI
Pick the right configuration for your model. All include fast NVMe storage.
- VRAM
- 16 GB
- CUDA cores
- 4,352
- FP4 compute
- 545 TOPS
- FP8 compute
- 272 TOPS
- Memory bandwidth
- 288 GB/s
- Processor
- 8 vCPU · 16 GB
- Storage
- NVMe
- VRAM
- 32 GB (2× 16)
- CUDA cores
- 8,704
- FP4 compute
- 1090 TOPS
- FP8 compute
- 544 TOPS
- Memory bandwidth
- 288 GB/s
- Processor
- 12 vCPU · 24 GB
- Storage
- NVMe
- VRAM
- 24 GB
- CUDA cores
- 8,960
- FP4 compute
- 770 TOPS
- FP8 compute
- 345 TOPS
- Memory bandwidth
- 432 GB/s
- Processor
- 8 vCPU · 16 GB
- Storage
- NVMe
- VRAM
- 48 GB (2× 24)
- CUDA cores
- 17,920
- FP4 compute
- 1540 TOPS
- FP8 compute
- 690 TOPS
- Memory bandwidth
- 432 GB/s
- Processor
- 12 vCPU · 24 GB
- Storage
- NVMe
Prices exclude VAT. Billing is hourly: you pay only for the hours the server runs. GPU servers are created from the panel like any other server, in the Timisoara location.
What you use them for
LLM inference
Run open-source language models (Llama, Mistral, Qwen) locally, with low latency and no per-token cost.
Fine-tuning
Train and fine-tune models on your own data, on a dedicated GPU, without sending anything to third parties.
Image generation
Stable Diffusion, Flux and other image models, with enough VRAM for high resolutions.
AI APIs and apps
Put your own AI endpoints into production, with full control over performance and cost.
Need GPU power for your AI project?
Create a GPU server straight from the panel, in a few minutes. Pay by the hour and stop it when you are done.
Create a GPU server