# GPU servers: getting started with your GPU VPS, VDS or dedicated server > First steps on a HyperDC GPU server: know shared vGPU from dedicated GPU, keep the right driver for NVIDIA, AMD or Intel, and check that the GPU works. Difficulty: Beginner HyperDC offers **GPU VPS** and **GPU VDS** with a shared vGPU or a dedicated GPU, and **GPU dedicated servers**, with GPUs from **Intel, NVIDIA or AMD**. Your package names its GPU. This guide explains the two GPU types, the driver each one needs, and how to check that the GPU answers. ## Before you start - Your server's login details from the welcome email, and SSH access (Linux) or Remote Desktop (Windows); see [connect to your Linux server](/guides/connect-to-linux-server-ssh). - The GPU and GPU type (shared vGPU or dedicated GPU) listed for your package on the service page or in the order. - Install all system updates first and reboot: `sudo apt update && sudo apt full-upgrade`. ## Shared vGPU or dedicated GPU | | Shared vGPU | Dedicated GPU | |---|---|---| | Memory | A fixed slice of the card's memory, reserved for your server from the moment the vGPU is created | All of the card's memory | | Compute | Time-sliced: the card's processing cycles are shared among the vGPUs on that card | The whole card, for your server only | | Driver inside your server | The **vGPU guest driver** that matches the host's vGPU software; preinstalled in our images | The vendor's standard **data-center driver** for the card | ## Step 1: Identify the GPU **Linux** ```bash lspci -nn | grep -iE "vga|3d|display" ls -l /dev/dri ``` **Windows** ```powershell Get-PnpDevice -Class Display | Format-Table FriendlyName, Status ``` **Verify:** the output names the GPU maker (NVIDIA, Advanced Micro Devices, or Intel), and on Windows the status is `OK`. If the GPU does not appear at all, open a ticket before installing anything. ## Step 2: Follow the driver path for your GPU | Your GPU | Driver path | |---|---| | **Shared vGPU** | Keep the vGPU guest driver that came with our image. If you ever need to reinstall it, use the driver version listed in your service details or provided by HyperDC support. | | **NVIDIA, dedicated GPU** | Install NVIDIA's data-center driver and CUDA: [install NVIDIA drivers and CUDA](/guides/install-nvidia-driver-cuda), or NVIDIA's [driver downloads](https://www.nvidia.com/en-us/drivers/) on Windows Server. | | **Intel, dedicated GPU** | Install Intel's Data Center GPU driver for your card and system, following [Intel's data-center GPU documentation](https://dgpu-docs.intel.com/). | | **AMD, dedicated GPU** | Install ROCm on Linux ([install AMD ROCm](/guides/install-amd-rocm)) or AMD's driver from [AMD's driver downloads](https://www.amd.com/en/support/download/drivers.html). | > **Warning** > > On a **shared vGPU**, the guest driver must be compatible with the vGPU software on the host. Do not install a standard consumer or desktop driver over it, and do not install another version: NVIDIA states that the vGPU then fails to load. On Windows, Device Manager shows the GPU with **Code 43**. To fix it, reinstall the matching guest driver, or reinstall the operating system image from the client area. ## Step 3: Check that the GPU works **NVIDIA** ```bash nvidia-smi ``` The table shows the GPU, the driver version and the GPU memory. On a shared vGPU, the memory shown is your vGPU's slice, not the whole card. **AMD** ```bash rocminfo | grep -i "Marketing Name:" amd-smi version ``` **Intel** ```bash clinfo | grep -i "device name" ``` ## Step 4: Run your workload - **Containers:** pass the GPU into Docker: [GPUs in Docker containers](/guides/gpu-in-docker-containers). - **Local AI models:** [install Ollama](/guides/install-ollama), [vLLM](/guides/install-vllm) or [ComfyUI](/guides/install-comfyui) for image generation. On a shared vGPU, choose models that fit your vGPU's memory. - **Monitoring:** keep an eye on temperature, memory and utilisation with `nvidia-smi -l 5` (NVIDIA) or `amd-smi` (AMD). ## Troubleshooting **`nvidia-smi: command not found`.** On a dedicated GPU, the NVIDIA utilities are not installed yet; see [install NVIDIA drivers and CUDA](/guides/install-nvidia-driver-cuda). On a shared vGPU, reinstall the matching guest driver or the operating system image. **`NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver`.** The kernel module is not loaded. Reboot after installing the driver; more causes in [GPU troubleshooting](/guides/gpu-troubleshooting). **The GPU shows Code 43 on Windows.** See the warning in step 2 and [GPU troubleshooting](/guides/gpu-troubleshooting). **The GPU is missing in `lspci`.** The GPU is not attached to the server. Open a ticket with the output of `lspci -nn`. ## Next steps - NVIDIA setup on a dedicated GPU: [install NVIDIA drivers and CUDA](/guides/install-nvidia-driver-cuda). - AMD setup: [install AMD ROCm](/guides/install-amd-rocm). - Something does not work: [GPU troubleshooting](/guides/gpu-troubleshooting). ## Frequently asked questions ### What is the difference between a shared vGPU and a dedicated GPU? A shared vGPU gets a fixed slice of a physical card's memory, reserved for your server, while the card's compute is time-sliced with the other servers on that card. A dedicated GPU is assigned to your server whole, with all of its memory and compute. ### Do I have to install the GPU driver? On a shared vGPU, no: our images ship with the matching vGPU guest driver. On a dedicated GPU, you install the vendor's standard data-center driver for the card, as described in this guide. ### Which operating system should I choose? One that your GPU software supports. NVIDIA's CUDA documentation lists current Ubuntu LTS and Debian releases, and AMD's ROCm documentation lists its supported distributions, which may lag behind the newest release. ### Can I use the GPU in Docker? Yes. NVIDIA GPUs use the NVIDIA Container Toolkit; AMD and Intel GPUs are passed to containers as devices. See our guide on GPUs in Docker. --- Source: \ Updated: 2026-10-16