[Issue]: Live Preview enabled significantly slows generation

[brknsoul]

Issue Description

I believe this is due to Live Preview slowing things down.

I'll attached the gifs I took;

Live preview display period = 1, method == approximate

Live preview display period = 0 (ie: disabled);

The slowdowns seems to coincide with spikes in vram usage (first spike = generation, second spike = detailer);

Version Platform Description

Python: version=3.10.11 platform=Windows bin="E:\sddev\venv\Scripts\python.exe"
venv="E:\sddev\venv"
Version: app=sd.next updated=2024-11-28 hash=75dd6219 branch=dev
url=https://github.com/vladmandic/automatic/tree/dev ui=dev
Repository latest available 6846f4e5d3e650d2a8dc8460906901086a350d27 2024-11-22T18:01:15Z
Platform: arch=AMD64 cpu=Intel64 Family 6 Model 151 Stepping 5, GenuineIntel system=Windows
release=Windows-10-10.0.19045-SP0 python=3.10.11 docker=False
Packages: venv=venv site=['venv', 'venv\\lib\\site-packages']
Args: ['--autolaunch', '--debug', '--uv']

Relevant log output

Command Prompt - astart.bat.txt

Backend

Diffusers

UI

Standard

Branch

Dev

Model

StableDiffusion 1.5

Acknowledgements

• I have read the above and searched for existing issues
• I confirm that this is classified correctly and its not an extension issue

[vladmandic]

i understand the issue, but are you saying this is NEW behavior and there was no slowdown previously?

[brknsoul]

Just installed master, no slow down.
Also, no GPU vram spikes, either.

[vladmandic]

argh. one of those that i cannot reproduce, likely because there are no slowdowns on high end gpus.

@brknsoul as a test, can you do one thing - in modules/sd_samplers_common.py, delete line:

with queue_lock:

[brknsoul]

My GPU is now a 4060Ti 16gb...
But I'll try.
No change.

[vladmandic]

can you try with the latest dev branch code?

[vladmandic]

resolved based on conversation in discord.

[brknsoul]

For future searches: The resolution was that I was using CUDAMallocAsync, instead of Native memory allocator.