Performance issue: system becomes sluggish on 4GB GPU laptops

[tsktech]

Summary

Portable Diffusion becomes very sluggish on a laptop with a 4GB NVIDIA GPU. During generation, the whole Windows system slows down badly enough that it feels practically unusable for normal laptop work.

Hardware / environment

• OS: Windows 11 laptop
• GPU: NVIDIA GeForce GTX 1650, 4GB VRAM
• Driver/CUDA shown locally: NVIDIA driver 596.36, CUDA 13.2
• Backend used: CUDA
• CPU threads tried: 4
• Model tested: CyberRealistic_V8_FP16.safetensors

Command used

C:\Users\Srika\OneDrive\Documents\Portable-Diffusion\app\backend\win\cuda\sd-cuda.exe --listen-port 8080 --model "C:\Users\Srika\OneDrive\Documents\Portable-Diffusion\app\models\CyberRealistic_V8_FP16.safetensors" --steps 20 --cfg-scale 7 --sampling-method euler_a --threads 4 --backend cuda0 --params-backend cpu --rng cuda --sampler-rng cuda

Actual behavior

• Generation starts, but the laptop becomes extremely sluggish.
• The UI/system responsiveness drops heavily.
• On a 4GB GPU laptop, the app does not feel practically usable.

Expected behavior

It would be helpful if Portable Diffusion either:

1. Provides a lower-memory / low-VRAM preset for 4GB GPUs, or
2. Warns users that 4GB GPUs may cause system-wide slowdown, or
3. Offers recommended launch settings for GTX 1650 / 4GB VRAM laptops.

Notes

This may not be a bug in the CUDA backend itself, but it would be useful to document recommended settings or minimum practical hardware requirements for 4GB GPU laptops.