ComfyUI and ComfyUI-MultiGPU
The custom node provides multi-GPU and virtual VRAM functionality, making it a complement that enhances the powerful, modular diffusion model GUI, API, and backend by extending its hardware utilization capabilities.
About ComfyUI
Comfy-Org/ComfyUI
The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface.
Supports diverse generative models across image, video, audio, and 3D modalities with intelligent memory management and GPU offloading for low-VRAM systems. The architecture uses an asynchronous queue system with incremental execution—only re-computing workflow nodes that have changed—and integrates LoRAs, ControlNets, and model merging capabilities. Extensible through custom nodes and external API providers, while maintaining fully offline operation for core functionality.
About ComfyUI-MultiGPU
pollockjj/ComfyUI-MultiGPU
This custom_node for ComfyUI adds one-click "Virtual VRAM" for any UNet and CLIP loader as well MultiGPU integration in WanVideoWrapper, managing the offload/Block Swap of layers to DRAM *or* VRAM to maximize the latent space of your card. Also includes nodes for directly loading entire components (UNet, CLIP, VAE) onto the device you choose
DisTorch2 uses a layer-distribution architecture to split model weights across CPU DRAM and multiple GPUs via byte-precise or ratio-based allocation, enabling users to offload static model components while reserving maximum VRAM for latent space computation. It supports universal `.safetensors` and GGUF model formats with three allocation modes: simple virtual VRAM sliders, expert byte/ratio specifications (e.g., `cuda:0,2.5gb;cpu,*`), and fraction-based distribution—compatible with all ComfyUI checkpoint, CLIP, VAE, ControlNet, and video generation loaders including WanVideoWrapper.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work