GPU Cloud Documentation

Use these guides to launch, connect to, manage, and troubleshoot GPU-backed Compute Nodes in MyAccount. The flow starts from Compute > Nodes with the GPU tab selected.

A GPU node is a regular Compute Node with an NVIDIA GPU attached. Apart from picking the GPU card first and the additional GPU fields on the plan screen, every other create-node, connect, and management step is the same as a regular node — see MyAccount Node for the canonical reference. For NVIDIA driver, CUDA, and other GPU-specific command-line issues, see Troubleshoot GPU Nodes.

Plan and Launch

Choose a GPU Card

Compare H100, A100, L40S, A40, A30, L4, V100, and T4 before launch.

Create a GPU Node

Pick a GPU card, then configure OS, plan, volume, and encryption.

Connect

Connect to a Linux GPU Node

SSH into a Linux GPU node and verify the driver with nvidia-smi.

Connect to a Windows GPU Node

RDP into a Windows GPU node and verify the NVIDIA driver.

Operate and Fix

Manage GPU Nodes

GPU-specific differences for monitoring, save image, plan upgrade, and reinstall.

Troubleshoot GPU Nodes

Fix NVIDIA driver, CUDA, GPU-not-detected, OOM, throttling, and container toolkit issues.

Guides

Run GPU Workloads in Docker

Install the NVIDIA Container Toolkit and run containers with --gpus all.

Bake and Reuse a GPU Image

What to bake into a GPU image and how to cut new-node boot to 3–6 minutes.

Serve LLM Inference

Run vLLM or TGI on a GPU node with card sizing and exposure patterns.