Skip to content

A whole graphics card for your model

The NVIDIA RTX PRO 6000 Blackwell Server Edition comes with 96 GB of graphics memory and more than 24,000 CUDA cores. In a GPU Seed you use one to eight of these cards exclusively, hosted in Germany.

What is inside the card

The professional card from NVIDIA's Blackwell generation, built for continuous operation in the data center.

65 of 96 GBgpt-oss-120b in graphics memory

Up to eight cards

A language model has to fit into graphics memory completely. Set the size of your model, the bars show how many cards it needs.

Well-known models

Fits on 1 card

96 GB graphics memory, 16 dedicated vCPUs, 1,60 € per hour
Configure with 1 card

Also on 1 card: gpt-oss-20b, Mistral Small 3.1, Gemma 3 27B, Qwen3 32B

OpenAI, 117 billion parameters (MoE), MXFP4. Estimates for the model weights, plus room for context.

65 of 96 GBgpt-oss-120b in graphics memory

The RTX PRO 6000 compared

Datasheet values of common graphics cards for AI and rendering. For large language models, graphics memory per card matters first.

RTX PRO 6000 in a GPU Seed Other card
RTX PRO 6000vs.H100PCIe, 80 GB HBM2e
96 GB
80 GB
+16 GB
RTX PRO 6000vs.L40S48 GB GDDR6
96 GB
48 GB
+48 GB
RTX PRO 6000vs.RTX 5090GeForce, 32 GB GDDR7
96 GB
32 GB
+64 GB
RTX PRO 6000vs.RTX 4090GeForce, 24 GB GDDR6X
96 GB
24 GB
+72 GB

Datasheet values are not measurements from our data center. The H100 with HBM memory and the RTX 5090 lead on memory bandwidth, the RTX PRO 6000 leads on graphics memory per card.

Manufacturer specifications from NVIDIA:RTX PRO 6000 Blackwell Server EditionH100L40SRTX 5090RTX 4090

How to start your model

Your GPU Seed starts with Ubuntu 24.04, graphics driver and CUDA 13.4. With root access you install your AI software and load the model, for example with Ollama or vLLM.

Go to the guides
curl -fsSL https://ollama.com/install.sh | sh
ollama pull gpt-oss:120b
curl -LsSf https://astral.sh/uv/install.sh | sh
source $HOME/.local/bin/env
DATA_DIR=~/.open-webui uvx --python 3.12 \
  open-webui@latest serve --port 3000
curl -LsSf https://astral.sh/uv/install.sh | sh
source $HOME/.local/bin/env
uv venv && source .venv/bin/activate
uv pip install vllm --torch-backend=auto
vllm serve openai/gpt-oss-120b --port 8000
apt install -y ffmpeg
ffmpeg -hwaccel cuda -i input.mov \
  -c:v av1_nvenc -preset p5 output.mp4
snap install blender --classic
blender -b scene.blend -E CYCLES -a \
  -- --cycles-device OPTIX

GPU Seeds and pricing

Open the configurator

Choose how many graphics cards your GPU Seed gets. Each configuration has an hourly price and a monthly cap.

GPUsVRAM CPU RAM StorageNetwork PriceConfigure
1×(Black Week Seed, available until November 30)96 GB16 vCPUs128 GB500 GB20 TB traffic
up to 10 Gbit/s
1,600962 € / Hour
max. 999,00 € / Month
1×96 GB16 vCPUs128 GB500 GB20 TB traffic
up to 10 Gbit/s
2,368100 € / Hour
max. 1.477,69 € / Month
2×(Black Week Seed, available until November 30)192 GB32 vCPUs256 GB1000 GB20 TB traffic
up to 10 Gbit/s
3,201923 € / Hour
max. 1.998,00 € / Month
2×192 GB32 vCPUs256 GB1000 GB20 TB traffic
up to 10 Gbit/s
4,736200 € / Hour
max. 2.955,39 € / Month
4×(Black Week Seed, available until November 30)384 GB64 vCPUs512 GB2000 GB20 TB traffic
up to 10 Gbit/s
6,403847 € / Hour
max. 3.996,00 € / Month
4×384 GB64 vCPUs512 GB2000 GB20 TB traffic
up to 10 Gbit/s
9,472400 € / Hour
max. 5.910,78 € / Month
8×(Black Week Seed, available until November 30)768 GB128 vCPUs1024 GB4000 GB20 TB traffic
up to 10 Gbit/s
12,807694 € / Hour
max. 7.992,00 € / Month
8×768 GB128 vCPUs1024 GB4000 GB20 TB traffic
up to 10 Gbit/s
18,944800 € / Hour
max. 11.821,56 € / Month
All prices incl. 19% VAT

Frequently asked questions about the RTX PRO 6000

What is a GPU Seed?

A GPU Seed is a cloud server with one or more dedicated graphics cards. You set up your own software on it and use it to run your AI models and applications.

Which GPU is available to me?

The NVIDIA RTX PRO 6000 Blackwell Server Edition (SE for short) with 96 GB of GDDR7 VRAM. You book one or more of them; each is assigned exclusively to your GPU Seed.

Can I also use the graphics card for video?

Yes. Each graphics card is fully available to your GPU Seed, with its full performance. That includes the card's four video encoders and four decoders (NVENC and NVDEC). They handle H.264, H.265 and AV1, and H.264 and H.265 in 4:2:2 as well.

Which AI models can I use?

In principle, any model that works with NVIDIA graphics cards and CUDA. That covers most common language and image models. How well a particular model runs depends on its size and on the number of graphics cards you book. The RTX PRO 6000 page shows how many cards a model needs.

Do I get a ready-made AI application?

GPU Seeds provide the infrastructure for your own software. You set up your models and applications yourself. For open-weights models such as Llama, Mistral, Qwen or gpt-oss, we are planning one-click apps that let you start a model with a single click.

Are graphics drivers and CUDA preinstalled?

Yes, if you choose our GPU image. It comes with graphics drivers and CUDA already set up, so you can install your AI software right away. The included versions are shown when you select the image. You can also choose any other image, such as Debian or Ubuntu, and install the drivers and CUDA yourself.

Can I use GPU Seeds for a single project or permanently?

Both. For a single project, you use the GPU Seed only for as long as you need it, billed by the hour. In ongoing operation, you protect your environment with snapshots and backups. The product terms in force at the time apply.

How does the selection work?

Mainly, you choose how many graphics cards your GPU Seed gets. vCPUs, memory and storage grow with the number of cards, and the pricing table shows every size and price. You then set the location, image and access, and add backups if you need them.

How quickly is my GPU Seed ready?

Your GPU Seed is ready to use within a few minutes. If a size is currently taken, you can book it once capacity frees up again.

When does billing end?

When you delete the GPU Seed. Shutting it down is not enough, because the graphics cards stay reserved for you.

What happens to my data when I delete the GPU Seed?

The system disk and snapshots are deleted with the GPU Seed, backups seven days later. Attached volumes keep all their data.

How are additional storage and traffic billed?

Every size comes with included traffic, the amount is shown in the pricing table. We look at traffic for your whole team rather than per GPU Seed: the included traffic of all Seeds is added up, prorated from the time of ordering to the end of the month. Only outbound traffic counts. Anything beyond that is billed per TB. You add more storage as a volume, billed by the hour for the size you book.

Ready for your own model?

One to eight NVIDIA RTX PRO 6000 Blackwell cards with 96 GB of graphics memory each, billed by the hour and hosted in Germany.