---
title: "GPU Node"
description: "A node built around an NVIDIA GPU card for training, inference, and other GPU work, created and managed from the same Nodes screen as any node."
content_type: "overview"
product: "gpu-node"
platform: "MyAccount"
url: "/myaccount/gpu-node/"
---

# GPU Node

A GPU node is a node built around an NVIDIA GPU card instead of, or alongside, plain CPU and
RAM: the same **Compute > Nodes** screen, the same create flow, the same lifecycle actions as
any node, with a card attached for work a CPU alone does slowly. Pick a GPU node for training,
inference, rendering, or another job that runs meaningfully faster on a GPU.

GPU plans belong to the **GDC** or **GDC3** family, both described in [Node families](/myaccount/node/concepts/node-families/), with cards ranging from the NVIDIA T4 and
L4 up to the A100. Hopper and later GPUs, the H100, H200 and B200, run only on [TIR](/tir/), our advanced AI/ML platform, ready-made for AI workloads. See [Choose a GPU card](/myaccount/gpu-node/concepts/choose-a-gpu-card/)
for what each one is built for.

## What is different from a regular node

Creating, connecting to, and managing a GPU node is mostly identical to a regular node,
covered in [Node](/myaccount/node/). This product's pages cover only what a GPU adds:

- [Create a GPU node](/myaccount/gpu-node/how-to/create-a-gpu-node/): pick a card first, then
  configure the rest the same way as any node.
- [Connect to a Linux GPU node](/myaccount/gpu-node/how-to/connect-to-a-linux-gpu-node/) or
  [a Windows GPU node](/myaccount/gpu-node/how-to/connect-to-a-windows-gpu-node/): the same SSH
  or RDP flow as any node, plus verifying the card with `nvidia-smi`.
- [Run GPU workloads in Docker](/myaccount/gpu-node/how-to/run-gpu-workloads-in-docker/): the
  recommended way to run a GPU job, with the driver on the host and everything else in a
  container.
- [Serve LLM inference](/myaccount/gpu-node/how-to/serve-llm-inference-on-a-gpu-node/): run
  vLLM or Hugging Face TGI behind an OpenAI-compatible API.
- [Save and reuse a GPU image](/myaccount/gpu-node/how-to/save-and-reuse-a-gpu-image/): bake
  the driver, toolkit, and your framework into an image so new nodes boot ready to work.

Everything else works exactly as on any node: power actions, security groups, VPC, volumes,
Accidental Protection, and recovery mode. See [Node](/myaccount/node/). Which actions a node
offers at any moment depends on its state and plan, and
[What each node action needs](/myaccount/node/troubleshooting/actions-and-lifecycle/#what-each-node-action-needs)
lists them.

For the NVIDIA driver, CUDA, and other GPU-specific issues on the command line, see
[Troubleshooting GPU Node](/myaccount/gpu-node/troubleshooting/gpu-node/).

## What you will see on a GPU node

A GPU node's details page adds its card, card count, GPU memory, and CUDA version to the
standard CPU, RAM, disk, and IOPS fields. The **Monitoring** tab adds graphs for GPU
utilization, GPU memory, GPU power, and GPU temperature, one set per card on a multi-card
plan, labeled **card_1**, **card_2**, and so on.

## Where to start with GPU nodes

[Get started with GPU Node](/myaccount/gpu-node/get-started/)