PSA: Use integrated graphics to save VRAM of nvidia GPU

All modern mobiles CPUs and many desktop ones too have integrated graphics. While iGPUs are useless for gaming and AI you can use them to run desktop apps and save precious VRAM for cuda tasks. Just connect display to motherboard output and done. You will be surprised how much VRAM modern apps eat, especially on Windows.

This is the end result with all desktop apps launched, dozen of browser tabs etc.
```
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.95.05 Driver Version: 580.95.05 CUDA Version: 13.0 |
+-----------------------------------------+------------------------+----------------------+
| 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. |
|=========================================+========================+======================|
| 0 NVIDIA GeForce RTX 5070 Ti Off | 00000000:01:00.0 Off | N/A |
| 0% 26C P8 8W / 300W | 15MiB / 16303MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| 0 N/A N/A 2064 G /usr/lib/xorg/Xorg 4MiB |
+-----------------------------------------------------------------------------------------+
```

I have appended `nvidia_drm.modeset=0` to GRUB_CMDLINE_LINUX_DEFAULT in `/etc/default/grub` but this should not be strictly necessary. Apparently there should be ridiculously complicated way to forbid Xorg from ever touching the GPU but I am fine with 4 Mb wasted.

https://redd.it/1psvxu8
@rStableDiffusion

Jib Mix ZIT - Out of Early Access

https://redd.it/1pt1tba
@rStableDiffusion

Is it just me or has the subreddit been over run with the same questions?

Between this account and my other account I’ve been with this subreddit for a while.

At the start this subreddit was filled with people asking real questions about things. Like tips or tricks for making unique workflows or understanding something. Recommend nodes to help with something particularly they’re trying to achieve. Maybe help trying to find a certain models after spending time searching and not able to find it. Or recommend videos or tutorials for something.

Now since Zimg or that what it seems like. Maybe Qwen it kinda started. Now it’s nothing but. “Best this, best that or best everything. How to make adult content this or that”..No actual real question I can try and answer.

The best question to me is” I’m new and don’t know anything and wanting to jump straight to using high end complex and advanced models or workflows without learning the very basics. So show me how to use it”


This could just be me. Or has anyone else that been doing this awhile have the same feeling?

https://redd.it/1pt1sop
@rStableDiffusion

We need a pin linking to the wiki (a guide to getting started), which should be updated. Too many redundant "how do I install a1111???" posts.

Every day there is at least one post which is something along the lines of

\- "Guys I can't install stable diffusion!!!"

\- "Guys why isn't a1111 working????? Something broke when I updated!!!"

\- "Guys I tried using *model from the last 1.5 years* and it makes this strange pattern??? btw it's stable diffusion"

\- "Guys I have an AMD GPU, what do I do????"

In the last 2 hours alone there were 2 posts like this. This sentiment also exists in the comments of unrelated posts, like people going "oh woe is me I don't understand Scratch, a shame Comfy is the only modern UI...".

The sub's wiki is a bit old, but all it needs is a small update linking to Stability Matrix, SDNext, Forge Classic Neo, etc., a big fat disclaimer to not use a1111 and that it's abandoned, cull the links to A1111/DirectML (which nukes performance), and add links to relevant ZLUDA/ROCm install guides - SDNext literally has docs for that, don't even need to include any explanation in the sub's wiki itself, just links. 5 minute change.

A pinned "read this before you make a new thread" post linking to such an updated wiki should hopefully inform people of how to properly get started, and reduce the number of these pointless posts that always have the same answer. Of course, there will always be people who refuse to read, but better than nothing.

https://redd.it/1pt5o1l
@rStableDiffusion

Why do I get better results with Qwen Image Edit 4 Step lora than original 20 step?

4 step takes less time and output is being better. Isn't more steps supposed to provide better image? I'm not familiar with this stuff but I thought slower/bigger/more steps would result in better results. But with 4 steps, it creates everything including text and the second image i uploaded accurately compared to 20 where text and the second image i asked for it to include gets distorted

https://redd.it/1pt6fdn
@rStableDiffusion