10 of 15 server configurations in stockInstant deliveryNo setup fees

2 GPU servers · 1 in stock

NVIDIA GPU servers for AI and rendering

A whole NVIDIA GPU that nobody else shares, on an EPYC host with ECC memory and NVMe: for inference, fine-tuning, Stable Diffusion, rendering and hardware video encoding.

  • From $435/mo
  • 1 of 2 ready to deploy
  • 5 locations
  • No setup fee
Dex carrying a large graphics card

GPU lineup

Pick your GPU server

Live stock and one monthly price. Open a server for its full specifications, or deploy it straight away.

Monthly prices, excl. VAT. Same price in every location where a server is offered.

Compare GPU servers side by side

Made for

NVIDIA GPUs for AI, rendering and transcoding

  • LLM inference with vLLM, Ollama or llama.cpp
  • Fine-tuning with LoRA and QLoRA
  • Stable Diffusion and ComfyUI pipelines
  • Blender rendering and NVENC transcoding

  • Dedicated GPU

    The whole card is yours: no time-slicing, no shared VRAM.

  • 4th-gen Tensor cores

    Ada Lovelace GPUs with FP8 support for efficient inference.

  • ECC end to end

    ECC VRAM on the GPU and ECC memory on the EPYC host.

  • Unmetered bandwidth

    Serve models and stream output without egress fees.

Solution guides

Sizing advice for your workload

Each guide recommends three servers, with a sizing table and setup tips.

Other server families

FAQ

GPU servers: questions

All questions
How large a model fits in 20 GB of VRAM?

As a rule of thumb, a 7–8B model quantized to 4 bits fits comfortably, with room for a long context, and 13B-class models fit in 4 or 8 bits. Larger models need the 48 GB L40S or a multi-GPU custom build. Always check your framework’s memory estimate.

Are NVIDIA drivers and CUDA installed?

You get a clean OS so you control versions. On Ubuntu, installing the driver and CUDA takes a few commands, and the NVIDIA Container Toolkit lets you run GPU workloads in Docker.

Can I use the GPU for video encoding?

Yes. Both GPUs include NVENC and NVDEC hardware video engines, supported by FFmpeg, for live transcoding and VOD pipelines.

The L40S is sold out. What are my options?

The RTX 4000 SFF Ada is in stock and covers most inference, image generation and rendering needs. For models that need 48 GB or more, ask for a custom quote and we will tell you the earliest date we can deliver.

Cleared for takeoff.

The NVIDIA RTX 4000 SFF Ada is ready to deploy at $435/mo. No setup fee, cancel anytime.

Deploy the RTX 4000 AdaGet a custom quote