Extensions/ComfyUI Multi-GPU Orchestrator
ComfyUI Extension

ComfyUI Multi-GPU Orchestrator

Multi GPU parallel queuing for ComfyUI. Keep all GPUs busy from one ComfyUI session by automatically dispatching workflows across every CUDA device.

By obsxrver·Created 4 months ago·Updated 4 days ago· 4
obsxrver/ComfyUI-MultiGPU-Orchestrator
Nodes—
On cloudLocal install
Stars4
Updated4 days ago
Readme

ComfyUI Multi-GPU Orchestrator

Use ALL the GPUs!

<p align="center"> <img width="48%" alt="ComfyUI-MultiGPU-Example 1" src="https://github.com/user-attachments/assets/c8bb0720-e7d3-4014-85b4-8f46461fb582" /> <img width="48%" alt="ComfyUI-MultiGPU-Example 2" src="https://github.com/user-attachments/assets/dad6aa42-3ce2-48a0-a2e2-c8a2b9a2d4d9" /> </p>

Run workflows across the CUDA GPUs on your system from one ComfyUI session. Queue jobs with the usual "► Run" button, and the orchestrator sends each job to the least-busy available GPU worker.

  • One ComfyUI instance controls all GPUs.
  • No workflow changes or added configuration required.
  • Queue progress, history, and generated media are available from the main UI.
  • Supports local and cloud-hosted systems with multiple CUDA GPUs.

Install

Clone this repository into ComfyUI/custom_nodes/:

cd ComfyUI/custom_nodes
git clone https://github.com/obsxrver/ComfyUI-MultiGPU-Orchestrator.git

Restart ComfyUI after installation. The extension automatically starts a worker for each visible CUDA GPU.

Usage

Open ComfyUI as usual, load a workflow, and click ► Run. Queue multiple jobs to keep multiple GPUs busy. Each job runs on one GPU; the extension does not split a single workflow across GPUs or combine their VRAM.

Use the normal queue controls to cancel individual jobs or cancel all jobs. Cancellation requires a ComfyUI version that supports the jobs cancellation API. If no GPU worker is available, jobs fall back to the main ComfyUI process.

Worker Settings

Open the MultiGPU sidebar to check worker status and configure:

  • Auto start at startup — start GPU workers automatically when ComfyUI starts.
  • Respawn failed workers — automatically restart workers after a failure.
  • Re-queue pending jobs on respawn — retry jobs last known to be running or waiting when a worker failed.

Workers stop when the main ComfyUI server shuts down or restarts.

Logs

In the Console, use Logs for the main ComfyUI process and GPU N for an individual worker. Worker logs are also saved to ComfyUI/logs/mgpu-workers/gpu-N.log and cleared when the orchestrator starts.