1VS.CO · ONE. VIRTUAL SERVICES CO.MPANY
Client area Talk to sales

GPU cloud

A whole GPU for training and inference.

Each server gets a full NVIDIA RTX PRO 6000 Blackwell through PCI passthrough, not a slice of one. Ubuntu 24.04 ships with CUDA and the NVIDIA Container Toolkit, so you can start working the moment you connect.

See plans and prices Talk to an engineer
  • From $2,398 / month
  • 96 GB GDDR7 VRAM
  • CUDA preinstalled
  • Full PCI passthrough
GDDR7 ECC memory on the GPU
96 GB
CUDA cores
24,064
FP32 TFLOPS
120
GPU memory bandwidth
1,597 GB/s

GPU Cloud Servers

One size, one full GPU

Choose a location

Same price in both locations. Your choice is added to the quote request.

GPU RTX PRO 6000

$2,398/ month

GPU
RTX PRO 6000 Blackwell
VRAM
96 GB GDDR7
vCPU
18
RAM
96 GB
Storage
900 GB NVMe
Get started
GPUNVIDIA RTX PRO 6000 Blackwell Server Edition
GPU memory96 GB GDDR7 with ECC
CUDA cores24,064
FP32 compute120 TFLOPS
Memory bandwidth1,597 GB/s
Tensor Cores5th generation
Video encoders4x NVENC (9th generation), AV1 and 4:2:2
GPU attachment1:1 PCI passthrough, no vGPU
ImageUbuntu 24.04 LTS with CUDA and NVIDIA Container Toolkit

Prices in USD, billed monthly unless stated otherwise. Need a different size? Ask for a custom configuration.

What you get

Built in, not bolted on.

The whole card is yours

The GPU is mapped 1:1 to your server through PCI passthrough. There is no vGPU slicing, so all VRAM, CUDA cores and memory bandwidth go to your workload.

CUDA ready on first boot

Servers deploy on Ubuntu 24.04 with the NVIDIA driver and CUDA toolkit installed. Connect over SSH and launch PyTorch, vLLM or any CUDA framework.

GPU containers built in

The NVIDIA Container Toolkit comes preinstalled, so Docker containers can use the GPU without extra setup.

Room for large models

At FP8, a 70B-parameter model needs about 70 GB, which leaves headroom in 96 GB of VRAM for context and batching.

Hardware video encoding

Four ninth-generation NVENC engines encode several 4K streams at once, with AV1 and 4:2:2 support.

Root access, your rules

Full root access and standard Linux tooling. Model weights, prompts and data stay on a server you control.

Use cases

What teams run on it.

  • 01

    Self-hosted LLM inference

    Serve open-weight models behind your own API, with prompts and outputs that never leave your server.

  • 02

    Fine-tuning and LoRA

    Adapt existing models with LoRA or full-parameter training on more VRAM than consumer cards offer.

  • 03

    Image and video generation

    Run Stable Diffusion and video generation models with memory to spare for large batches and high resolutions.

  • 04

    Rendering and simulation

    Render Blender or Cinema 4D scenes overnight, or run CFD and molecular dynamics on a single node without building a cluster.

How it works

Three steps to running.

  1. 01

    Pick a region

    Choose Istanbul or Nuremberg for your GPU server.

  2. 02

    Order

    The server deploys on Ubuntu 24.04 with the CUDA driver, CUDA toolkit and NVIDIA Container Toolkit already in place.

  3. 03

    Connect and run

    SSH in and start your framework, container or notebook straight away.

Questions

GPU Cloud Servers, answered.

Is the GPU shared with other customers?

No. The full RTX PRO 6000 is attached to your server through 1:1 PCI passthrough, with no vGPU slicing.

Can I train models, or is it only for inference?

Both. With 96 GB of VRAM and fifth-generation Tensor Cores, the card handles inference, fine-tuning and training. Adapting existing models is where it fits best.

Will a 70B model fit?

Yes, at FP8 precision. The weights take roughly 70 GB at one byte per parameter, leaving room in 96 GB for context and inference overhead.

Which operating system does it run?

GPU servers deploy on Ubuntu 24.04 LTS with the NVIDIA driver, CUDA toolkit and NVIDIA Container Toolkit already installed.

What if I need more than one GPU?

A GPU cloud server has one card. For multi-GPU work, our GPU dedicated servers offer up to four RTX PRO 6000 cards in one machine.

Is the card certified for professional 3D software?

Yes. The RTX PRO 6000 is ISV-certified for Maya, Houdini and Cinema 4D, so production rendering pipelines run on tested hardware.

More cloud

Often used together.

Cloud VPS

Virtual machines in four families, from shared-CPU starters to dedicated, memory-heavy servers.

From $12 / month

Explore

GPU Dedicated Servers

Bare-metal servers with NVIDIA RTX PRO Blackwell GPUs, from one 24 GB card to four 96 GB cards.

From $477.22 / month

Explore

Managed Kubernetes

Kubernetes clusters with a managed control plane. You pick the worker nodes; we run the rest.

From $24 / month

Explore

Talk to an engineer

Not sure which size fits?

Tell us what you want to run. An engineer suggests a configuration and sends you a written price.

Request a quote +1 850 605 09 19