GPU servers are not orderable yet. We are preparing servers with an NVIDIA Tesla P40 (24 GB VRAM) passed through to each one. Contact us and we will tell you when you can deploy one.
Each GPU will be passed through to a single server over PCIe.
Workloads a single 24 GB GPU is suited to, once ordering opens
Fine-tune and train models with PyTorch, TensorFlow or JAX. A single 24 GB Tesla P40 suits small models and LoRA fine-tuning rather than large-scale training.
Ask about availability →Serve LLMs, image generation, speech recognition and computer vision models from one GPU.
Ask about availability →Transcode video with NVENC hardware encoding and render 3D scenes with Blender Cycles.
Ask about availability →Run molecular dynamics, simulations and other CUDA or OpenCL workloads that work in single precision.
Ask about availability →Speed up array math and gradient boosting with CuPy, XGBoost and PyTorch on the GPU.
Ask about availability →Build and test AI applications with Jupyter notebooks, VS Code remote and your own CUDA development environment.
Ask about availability →PCIe passthrough gives your server the whole GPU, with close to native performance.
GPU servers cannot be ordered from the panel yet. Contact us with your workload and we will reply when capacity is ready.
VPS.org plans are billed hourly, capped at 672 hours a month, with no contract. GPU pricing will be published when ordering opens.
Install any framework, driver version or CUDA toolkit. SSH access from day one.
Guides for popular AI/ML frameworks, creative tools and GPU-accelerated applications
Everything you need to know about GPU VPS hosting
A GPU VPS is a virtual server with a physical GPU passed through to it, for work such as AI inference, model fine-tuning, video transcoding and rendering. VPS.org GPU servers are not orderable yet.
None can be ordered yet. The first GPU servers we are preparing use NVIDIA Tesla P40 cards with 24 GB of VRAM. Contact us to be notified when they open.
With PCIe passthrough, a whole physical GPU is assigned to one virtual server. Your software uses the card directly through the standard NVIDIA driver, with close to native performance, and the GPU is not shared with other customers.
No. You install the NVIDIA driver and the CUDA toolkit version you need yourself, with full root access. The Tesla P40 is a Pascal GPU (compute capability 6.1), so check that your framework version still supports it.
Not at the moment. A GPU server will have one GPU. Contact us if your workload needs more.
Model inference, fine-tuning of small and quantized models, computer vision, speech recognition, video transcoding, 3D rendering and scientific computing that runs in single precision.
GPU servers are not orderable yet, so there is nothing to pay today. Standard VPS plans are billed hourly, capped at 672 hours a month; GPU pricing will be published when ordering opens.
Not in place. GPU servers run on separate GPU nodes, so moving to one means deploying a new server and copying your data across.
No. VPS.org does not include denial-of-service mitigation on any plan. If you need it, put public services behind a proxy or mitigation service.
Supported operating systems will be confirmed when ordering opens. Ubuntu LTS releases have the most mature NVIDIA driver packages.
There is no contract on VPS.org plans: you can destroy a server at any time and billing stops. GPU terms will be published when ordering opens.
Contact us through the contact page with your workload and the GPU memory you need, and we will let you know when GPU servers can be ordered.
GPU servers are not orderable yet. Tell us about your workload and we will let you know when they are.