第六十一篇-ComfyUI+V100-32G+GGUF+运行Flux Schnell GGUF

环境

系统：CentOS-7 CPU : E5-2680V4 14核28线程 内存：DDR4 2133 32G * 2 显卡：Tesla V100-32G【PG503】 (水冷) 驱动: 535 CUDA: 12.2 ComfyUI version: 0.4.0 ComfyUI frontend version: 1.34.8

系统软件信息

系统信息 OS linux Python Version 3.12.12 | packaged by Anaconda, Inc. | (main, Oct 21 2025, 20:16:04) [GCC 11.2.0] Embedded Python false Pytorch Version 2.9.1+cu128 Arguments main.py --listen --port 8188 --cuda-malloc --lowvram RAM Total 62.68 GB RAM Free 60.25 GB

启动

python main.py --listen --port8188--cuda-malloc --lowvram

参考

基于ComfyUI的Flux Schnell案例修改模型为GGUF加载器 [第五十九篇-ComfyUI+V100-32G+运行Flux Schnell-CSDN博客](https://blog.csdn.net/hai4321/article/details/155953374)

ComfyUI安装GGUF支持

进入你看着ComfyUI目录的custom_nodes cd ComfyUI/custom_nodes 克隆代码 git clone https://github.com/city96/ComfyUI-GGUF 安装依赖 pip install -r requirements.txt 重启ComfyUI

下载GGUF模型

https://hf-mirror.com/city96/FLUX.1-schnell-gguf/resolve/main/flux1-schnell-Q4_K_S.gguf?download=true
放入ComfyUI/models/unet文件夹中

调整模型加载器

删除Setp1 UNet加载器

添加【节点库】–》【UnetLoader(GGUF)】-》【选择flux1-schnell-Q4_K_S.gguf】

Flux Schnell完整版文生图

保存工作流

Ctrl+S

运行结果

第一次时间长一点

参数

1024*1024

时间

gguf qtypes: F32(468), Q4_K(304), F16(4)model weight dtype torch.float16, manual cast: None model_type FLOW Requested to load Flux loaded completely;30387.70MB usable,6595.58MB loaded, full load: True100%|███████████████████████████████████████████████████████████████████████████████████████|4/4[00:06<00:00,1.69s/it]Prompt executedin11.09seconds got prompt100%|███████████████████████████████████████████████████████████|4/4[00:06<00:00,1.70s/it]Prompt executedin7.59seconds got prompt100%|███████████████████████████████████████████████████████████|4/4[00:06<00:00,1.70s/it]Prompt executedin7.61seconds

GPU

Tue Dec1621:42:542025+---------------------------------------------------------------------------------------+|NVIDIA-SMI535.129.03 Driver Version:535.129.03 CUDA Version:12.2||-----------------------------------------+----------------------+----------------------+|GPU Name Persistence-M|Bus-Id Disp.A|Volatile Uncorr. ECC||Fan Temp Perf Pwr:Usage/Cap|Memory-Usage|GPU-Util Compute M.||||MIG M.||=========================================+======================+======================||0Tesla PG503-216 On|00000000:04:00.0 Off|0||N/A 21C P0 36W / 250W|7184MiB / 32768MiB|0% Default||||N/A|+-----------------------------------------+----------------------+----------------------+

总结

1.GPU 占用7-8G 2.GPU 100% 3.7 秒左右一张1024*1024 4.还是挺好用的，GPU内存占用只有7G多，比fp16少很多