---
title: "ComfyUI on NVIDIA RTX 5090 & 5080 (CUDA 12.8 & FP8) | ComfyUI Windows"
description: "ComfyUI on NVIDIA RTX 5090 & 5080 (CUDA 12.8 & FP8) — practical answer and step-by-step Windows setup for \"comfyui rtx 5090 5080 setup\" in ComfyUI. Download the desktop or portable build at /download."
canonical: https://comfy.ink/guides/comfyui-rtx-5090-5080-blackwell-setup
updated: 2026-10-17
---

# ComfyUI on NVIDIA RTX 5090 & 5080 (CUDA 12.8 & FP8)

Direct Windows guide for "comfyui rtx 5090 5080 setup": folder paths, node wiring, and VRAM settings in ComfyUI.

## Facts (2026-10-17)

- **Search Topic:** comfyui rtx 5090 5080 setup
- **Category:** Guides
- **OS Support:** Windows 10 & Windows 11 x64
- **Download:** /download

## How "comfyui rtx 5090 5080 setup" works in ComfyUI on Windows

Users searching for **comfyui rtx 5090 5080 setup** want a straightforward, working setup on Windows without broken Python dependencies or wasted VRAM. Modern diffusion models (Flux.1, SDXL, Hunyuan Video, and Wan 2.1) scale across 8 GB to 24 GB Windows GPUs when you pair the right loader node with FP8 or GGUF quantization.

ComfyUI separates model loading ([Load Checkpoint](https://comfy.ink/nodes/load-checkpoint) / [GGUF Loader](https://comfy.ink/nodes/gguf-loader-nodes)), text conditioning ([CLIP Text Encode](https://comfy.ink/nodes/clip-text-encode)), denoising ([KSampler](https://comfy.ink/nodes/ksampler)), and pixel decoding ([VAE Decode](https://comfy.ink/nodes/vae-decode-encode)) into explicit nodes so you have full visibility into every step.

## Step-by-step checklist

1. Grab the latest Windows desktop or portable installer from [/download](https://comfy.ink/download) and launch [ComfyUI Studio](https://comfy.ink/studio).
2. Place your `.safetensors` or `.gguf` weights into the matching subfolder under [models/](https://comfy.ink/models) (or link an existing folder via [extra_model_paths.yaml](https://comfy.ink/install/extra-model-paths-yaml-setup)).
3. Open [ComfyUI-Manager](https://comfy.ink/nodes/comfyui-manager) to install any missing custom nodes if you imported a community JSON or PNG workflow.
4. If VRAM is 8 GB–12 GB, use FP8 or GGUF Q4 weights and enable `VAE Decode (Tiled)` before queuing your prompt.

## FAQ

### What is the fastest way to set up "comfyui rtx 5090 5080 setup" on Windows?

Install the Windows build from [/download](https://comfy.ink/download), verify your weights are inside the matching `models/` folder, and press **R** on the canvas to refresh node dropdowns.

### Does this require a paid cloud account or internet connection?

No. Once you have downloaded ComfyUI and your chosen model weights, everything runs locally on your own Windows GPU offline.

## Related

- [ComfyUI Guides](https://comfy.ink/guides)
- [ComfyUI Desktop Windows Download](https://comfy.ink/desktop/comfyui-desktop-windows-download)
- [ComfyUI VRAM Requirements for SDXL, Flux & Video (6GB–24GB)](https://comfy.ink/guides/comfyui-vram-requirements-guide)
- [ComfyUI-Manager](https://comfy.ink/nodes/comfyui-manager)
- [Complete Flux.1 Dev & Schnell Setup Guide in ComfyUI](https://comfy.ink/guides/comfyui-flux-complete-setup-guide)

## Download

Free desktop app: [Download Comfy](https://comfy.ink/download)

Markdown mirror: https://comfy.ink/guides/comfyui-rtx-5090-5080-blackwell-setup.md