Home Creators Posts Import Register Favorites Logout

Content

BF16, GGUF Q8, FP8 Scaled and NVFP4 precisions compared for speed and quality in ComfyUI with CUDA 13; NVFP4 ran 87% faster than GGUF Q8 on Z Image Turbo.

Tested with CUDA 13 and Sage Attention on Z Image Turbo, FLUX Dev, FLUX SRPO, FLUX Kontext and FLUX 2, with the difference in quality shown side by side. Also introduces our NVFP4 model quantizer and FP8 Scaled quantization apps.

Full tutorial link > https://www.youtube.com/watch?v=XDzspWgnzxI

Compared Quality and Speed Difference (with CUDA 13 & Sage Attention) of BF16 vs GGUF Q8 vs FP8 Scaled vs NVFP4 for Z Image Turbo, FLUX Dev, FLUX SRPO, FLUX Kontext, FLUX 2

BF16 vs GGUF, FP8 Scaled, NVFP4 Speed & Quality Compared + ComfyUI CUDA 13 Gains + FLUX 2 Klein 9B

Video Info

It was always wondered how much quality and speed difference exists between BF16, GGUF, FP8 Scaled and NVFP4 precisions. In this tutorial I have compared all these precision and quantization variants for both speed and quality. The results are pretty surprising. Moreover, we have developed and published NVFP4 model quant generator app and FP8 Scaled quant generator apps. The links of the apps are below if you want to use them. Furthermore, upgrading ComfyUI to CUDA 13 with properly compiled libraries is now very much recommended. We have observed some noticeable performance gains with CUDA 13. So for both SwarmUI and ComfyUI solo users, CUDA 13 ComfyUI is now recommended.

📂 Resources & Links:

⏱️ Video Chapters:

  • 00:00:00 Introduction: GGUF Q8 vs NVFP4 vs BF16 vs FP8 Precision Comparison

  • 00:00:38 FP8 Quantization & New NVFP4 Model Quantizer App in Musubi Trainer

  • 00:01:08 The New FLUX SRPO Mixed NVFP4 Model & FLUX 2 Klein 9B Announcement

  • 00:01:56 Speed Comparison Setup: ComfyUI CUDA 13 & Compiled Libraries

  • 00:02:41 Z Image Turbo Speed Test: GGUF Q8 vs NVFP4 (87% Faster)

  • 00:03:09 Z Image Turbo Speed Test: BF16 vs FP8 Scaled vs GGUF Improvements

  • 00:03:32 Installing & Using Image Comparison Slider Tool for Quality Check

  • 00:03:55 Z Image Turbo Quality: BF16 vs GGUF Q8 vs FP8 Scaled

  • 00:04:13 Z Image Turbo Quality: NVFP4 Degradation Analysis

  • 00:04:27 FLUX 2 Dev Speed Test: GGUF Q8 vs NVFP4 (100% Faster)

  • 00:04:43 FLUX 2 Dev Speed Test: FP8 Scaled vs BF16 Performance

  • 00:05:12 FLUX 2 Dev Quality: BF16 vs GGUF Q8 vs Mixed FP8 Scaled

  • 00:05:38 FLUX 2 Dev Quality: NVFP4 Mixed Precision Analysis

  • 00:05:54 Benchmark Settings: 2048px Resolution & Quality 1 Preset Details

  • 00:06:25 FLUX 1 Dev Speed Test: GGUF Q8 vs NVFP4 (118% Faster)

  • 00:07:21 FLUX 1 Dev Speed Test: BF16 & FP8 Scaled Performance Stats

  • 00:07:42 FLUX 1 Dev Quality: BF16 vs GGUF Q8 vs FP8 Scaled

  • 00:07:55 FLUX 1 Dev Quality: NVFP4 Visual Degradation Review

  • 00:08:06 FLUX 1 Kontext Dev: Model Intro & Outpainting Tutorial Reference

  • 00:08:40 FLUX 1 Kontext Dev Speed: GGUF Q8 vs NVFP4 (93% Faster)

  • 00:08:59 FLUX 1 Kontext Dev Speed: BF16 & FP8 Scaled Comparisons

  • 00:09:12 FLUX 1 Kontext Dev Quality: Original vs Edited Image (Hair Change)

  • 00:09:36 FLUX 1 Kontext Dev Quality: BF16 vs GGUF Q8 vs FP8 Scaled

  • 00:09:51 How to Use SwarmUI Unified Model Downloader & Bundles

  • 00:10:36 Downloading Models via URL from CivitAI & Hugging Face to Cloud

  • 00:11:45 SECourses Musubi Trainer: Creating Custom FP8 Quantized Models

  • 00:12:44 The New FLUX SRPO NVFP4 Mixed Precision Model Overview

  • 00:13:15 Live Demo: FLUX SRPO NVFP4 Speed Test on RTX 5090 (5.7s)

  • 00:13:52 VRAM Usage Analysis: NVFP4 on RTX 5090 (14GB Usage)

  • 00:14:16 Live Comparison: BF16 Speed & VRAM Test on RTX 5090

  • 00:15:15 Troubleshooting: Fixing Low RAM/VRAM Issues with Arguments

  • 00:16:25 Why You Should Upgrade to ComfyUI CUDA 13 Version

  • 00:16:51 SimplePod AI: Updated Instructions & Template Setup

  • 00:17:29 RTX 6000 Blackwell Fix & nvitop Utilization Verification

  • 00:18:18 Conclusion, Contact Info & Support Channels

  • In this video, you will learn:

  • Speed differences between GGUF Q8, NVFP4, BF16, and FP8.

  • Visual quality analysis using the Image Comparison Slider.

  • How to use the new NVFP4 and FP8 Quantizer tools.

  • How to fix Low VRAM/RAM issues with specific arguments.

  • Performance benchmarks on RTX 5090 and RTX 6000.

Comparison Screenshots

01

Related posts

All posts in this topic: Research, Comparisons and Experiments · New to SECourses? Start Here: Best and Most Popular SECourses Posts

Files

BF16 vs GGUF, FP8 Scaled, NVFP4 Speed & Quality Compared + ComfyUI CUDA 13 Gains + FLUX 2 Klein 9B

It was always wondered how much quality and speed difference exists between BF16, GGUF, FP8 Scaled and NVFP4 precisions. In this tutorial I have compared all these precision and quantization variants for both speed and quality. The results are pretty surprising. Moreover, we have developed and published NVFP4 model quant generator app and FP8 Scaled quant generator apps. The links of the apps are below if you want to use them. Furthermore, upgrading ComfyUI to CUDA 13 with properly compiled libraries is now very much recommended. We have observed some noticeable performance gains with CUDA 13. So for both SwarmUI and ComfyUI solo users, CUDA 13 ComfyUI is now recommended. 📂 Resources & Links: 📥 Download ComfyUI CUDA 13 Installer: [ https://www.patreon.com/posts/ComfyUI-Installers-105023709 ] 📥 SwarmUI & ComfyUI Unified Model Downloader: [ https://www.patreon.com/posts/SwarmUI-Install-Download-Models-Presets-114517862 ] 🤖 NVFP4 Model Quantizer App: [ https://www.patreon.com/posts/nvfp4-quantizer-app-148217625 ] 🤖 SECourses Musubi Trainer (FP8 Scaled Quantization App): [ https://www.patreon.com/posts/nvfp4-quantizer-app-148217625 ] 🛠️ Image Comparison Slider Tool: [ https://www.patreon.com/posts/image-video-comparison-slider-app-133935178 ] ☁️ SimplePod AI: [ https://simplepod.ai/ref?user=secourses ] New Model FLUX 2 Klein 9B: [ https://huggingface.co/black-forest-labs/FLUX.2-klein-9B ] 🎥 FLUX 1 Kontext Dev Tutorial (inpaint - outpaint - image fix): [ https://youtu.be/XWzZ2wnzNuQ ] 🎥 Previous ComfyUI Installation Tutorial: [ https://youtu.be/yOj9PYq3XYM ] How to Use SwarmUI Presets & Workflows in ComfyUI + Custom Model Paths Setup for ComfyUI & SwarmUI Tutorial: [ https://youtu.be/EqFilBM3i7s ] SECourses Discord Channel for 7/24 Support: [ https://discord.com/invite/software-engineering-courses-secourses-772774097734074388 ] SECourses Musubi Tuner Tutorial: [ https://youtu.be/DPX3eBTuO_Y ] NVIDIA NVFP4 Blog Post to learn More: [ https://developer.nvidia.com/blog/introducing-nvfp4-for-efficient-and-accurate-low-precision-inference/ ] ⏱️ Video Chapters: 0:00 Introduction: GGUF Q8 vs NVFP4 vs BF16 vs FP8 Precision Comparison 0:38 FP8 Quantization & New NVFP4 Model Quantizer App in Musubi Trainer 1:08 The New FLUX SRPO Mixed NVFP4 Model & FLUX 2 Klein 9B Announcement 1:56 Speed Comparison Setup: ComfyUI CUDA 13 & Compiled Libraries 2:41 Z Image Turbo Speed Test: GGUF Q8 vs NVFP4 (87% Faster) 3:09 Z Image Turbo Speed Test: BF16 vs FP8 Scaled vs GGUF Improvements 3:32 Installing & Using Image Comparison Slider Tool for Quality Check 3:55 Z Image Turbo Quality: BF16 vs GGUF Q8 vs FP8 Scaled 4:13 Z Image Turbo Quality: NVFP4 Degradation Analysis 4:27 FLUX 2 Dev Speed Test: GGUF Q8 vs NVFP4 (100% Faster) 4:43 FLUX 2 Dev Speed Test: FP8 Scaled vs BF16 Performance 5:12 FLUX 2 Dev Quality: BF16 vs GGUF Q8 vs Mixed FP8 Scaled 5:38 FLUX 2 Dev Quality: NVFP4 Mixed Precision Analysis 5:54 Benchmark Settings: 2048px Resolution & Quality 1 Preset Details 6:25 FLUX 1 Dev Speed Test: GGUF Q8 vs NVFP4 (118% Faster) 7:21 FLUX 1 Dev Speed Test: BF16 & FP8 Scaled Performance Stats 7:42 FLUX 1 Dev Quality: BF16 vs GGUF Q8 vs FP8 Scaled 7:55 FLUX 1 Dev Quality: NVFP4 Visual Degradation Review 8:06 FLUX 1 Kontext Dev: Model Intro & Outpainting Tutorial Reference 8:40 FLUX 1 Kontext Dev Speed: GGUF Q8 vs NVFP4 (93% Faster) 8:59 FLUX 1 Kontext Dev Speed: BF16 & FP8 Scaled Comparisons 9:12 FLUX 1 Kontext Dev Quality: Original vs Edited Image (Hair Change) 9:36 FLUX 1 Kontext Dev Quality: BF16 vs GGUF Q8 vs FP8 Scaled 9:51 How to Use SwarmUI Unified Model Downloader & Bundles 10:36 Downloading Models via URL from CivitAI & Hugging Face to Cloud 11:45 SECourses Musubi Trainer: Creating Custom FP8 Quantized Models 12:44 The New FLUX SRPO NVFP4 Mixed Precision Model Overview 13:15 Live Demo: FLUX SRPO NVFP4 Speed Test on RTX 5090 (5.7s) 13:52 VRAM Usage Analysis: NVFP4 on RTX 5090 (14GB Usage) 14:16 Live Comparison: BF16 Speed & VRAM Test on RTX 5090 15:15 Troubleshooting: Fixing Low RAM/VRAM Issues with Arguments 16:25 Why You Should Upgrade to ComfyUI CUDA 13 Version 16:51 SimplePod AI: Updated Instructions & Template Setup 17:29 RTX 6000 Blackwell Fix & nvitop Utilization Verification 18:18 Conclusion, Contact Info & Support Channels In this video, you will learn: Speed differences between GGUF Q8, NVFP4, BF16, and FP8. Visual quality analysis using the Image Comparison Slider. How to use the new NVFP4 and FP8 Quantizer tools. How to fix Low VRAM/RAM issues with specific arguments. Performance benchmarks on RTX 5090 and RTX 6000.

Comments

Julius

Hi, since last upadate ComfyUI demands package ftfy, peft, pyloudnorm, in lib folder, any idea how to install these packages I ve tried to use requirements.txt, but package are always missing.

SECourses

i am checking the issue. when do you get this error exactly and on windows=?

Julius

Yes, I'm on Windows. When I launch SwarmUI, this error appears in yellow in the command prompt window. This warning doesn't prevent me from launching SwarmUI, but I imagine it has an impact.

SECourses

are you python 3.10? I will do a fresh install test

Julius

Yes when I installed swarmUI I have chossen python 3.10 as you advised.

SECourses

ok i am making a fresh comfyui and swarmui installation with python 3.10 to test

Julius

Thank you for your message, but the image you posted on ibb.co is no longer available.

Julius

Hello, any idea to resolve this error? "ComfyUI execution error: The size of tensor a (48) must match the size of tensor b (16) at non-singleton dimension 1"

SECourses

well impossible to know with a guess. are you using one of my presets? give me more info please

Julius

ComfyUI is very demanding and temperamental; it can freeze on the slightest change to the settings or on using a model that isn't in the presets. It's very frustrating. I wanted to test the wan2.2_i2v-rapid-mega-aio model, but it wouldn't accept it!

Julius

But to remain positive, I must admit I'm very impressed with the LTX2 i2v; for me, it's the best. Very fast, very realistic, very creative, and very responsive to prompts. The model accepts my 1500x844 resolution and 191 frames without any issues. Thank you for this model.

MuneRoons

I'm currently debating between a 24gb 3090 and a 16gb 5070ti. With all these new tools coming out for Blackwell hardware, which would you recommend I go with?