PC shuts down randomly during img2img generation

[huangfanglong]

This sometimes happens when I use img2img with Ultimate SD upscaler to upscale an image 2x from 1024x1024). My PC would just shut down randomly during process.

It works most of the time but sometimes it just shut down. Prior to the shutdown, I notice no slow downs or hangs or freeze whatsoever. Screen just goes black as I was doing my thing and it's processing in the background, followed by my keyboard RGB going out then PC shuts off.

PC specs:
AMD Ryzen 9 5900X
AMD Radeon 6900 XT - Using ZLUDA.
32GBs of RAM
850w PSU

I followed the https://github.com/vladmandic/automatic/wiki/ZLUDA guide to use ZLUDA and it's been working phenomenal minus this one occasional issue. So far has happened 3 or 4 times after 24 hours of use (not continuous, just total since I got it installed).

Anyone has any ideas?

[vladmandic]

could be anything, but I'd start with monitor your temps - and not just overall, but also peaks, check out cpuz and gpuz apps.
also, check event logs for any events raised just prior to shutdown.

[huangfanglong]

I'll keep an eye out for that. Been looking at my GPU temp periodically when I'm generating stuff and it doesn't seem to go beyond 80.

Also, as an unrelated question, do you know why sometimes when I take a txt2image generated image to img2img and starts the upscale process, it outputs the following error after I click generate and do nothing:

───────────────────────────────────────── Traceback (most recent call last) ──────────────────────────────────────────╮
│ A:\Stable Diffusion\automatic\modules\call_queue.py:31 in f                                                          │
│                                                                                                                      │
│   30 │   │   │   try:                                                                                                │
│ ❱ 31 │   │   │   │   res = func(*args, **kwargs)                                                                     │
│   32 │   │   │   │   progress.record_results(id_task, res)                                                           │
│                                                                                                                      │
│ A:\Stable Diffusion\automatic\modules\img2img.py:265 in img2img                                                      │
│                                                                                                                      │
│   264 │   else:                                                                                                      │
│ ❱ 265 │   │   processed = modules.scripts.scripts_img2img.run(p, *args)                                              │
│   266 │   │   if processed is None:                                                                                  │
│                                                                                                                      │
│ A:\Stable Diffusion\automatic\modules\scripts.py:483 in run                                                          │
│                                                                                                                      │
│   482 │   │   parsed = p.per_script_args.get(script.title(), args[script.args_from:script.args_to])                  │
│ ❱ 483 │   │   processed = script.run(p, *parsed)                                                                     │
│   484 │   │   s.record(script.title())                                                                               │
│                                                                                                                      │
│ A:\Stable Diffusion\automatic\extensions\ultimate-upscale-for-automatic1111\scripts\ultimate-upscale.py:565 in run   │
│                                                                                                                      │
│   564 │   │   upscaler.add_extra_info()                                                                              │
│ ❱ 565 │   │   upscaler.process()                                                                                     │
│   566 │   │   result_images = upscaler.result_images                                                                 │
│                                                                                                                      │
│ A:\Stable Diffusion\automatic\extensions\ultimate-upscale-for-automatic1111\scripts\ultimate-upscale.py:138 in proce │
│                                                                                                                      │
│   137 │   │   if self.redraw.enabled:                                                                                │
│ ❱ 138 │   │   │   self.image = self.redraw.start(self.p, self.image, self.rows, self.cols)                           │
│   139 │   │   │   self.initial_info = self.redraw.initial_info                                                       │
│                                                                                                                      │
│                                               ... 14 frames hidden ...                                               │
│                                                                                                                      │
│ A:\Stable Diffusion\automatic\venv\lib\site-packages\torch\nn\modules\module.py:1541 in _call_impl                   │
│                                                                                                                      │
│   1540 │   │   │   │   or _global_forward_hooks or _global_forward_pre_hooks):                                       │
│ ❱ 1541 │   │   │   return forward_call(*args, **kwargs)                                                              │
│   1542                                                                                                               │
│                                                                                                                      │
│ A:\Stable Diffusion\automatic\venv\lib\site-packages\diffusers\models\downsampling.py:147 in forward                 │
│                                                                                                                      │
│   146 │   │                                                                                                          │
│ ❱ 147 │   │   hidden_states = self.conv(hidden_states)                                                               │
│   148                                                                                                                │
│                                                                                                                      │
│ A:\Stable Diffusion\automatic\venv\lib\site-packages\torch\nn\modules\module.py:1532 in _wrapped_call_impl           │
│                                                                                                                      │
│   1531 │   │   else:                                                                                                 │
│ ❱ 1532 │   │   │   return self._call_impl(*args, **kwargs)                                                           │
│   1533                                                                                                               │
│                                                                                                                      │
│ A:\Stable Diffusion\automatic\venv\lib\site-packages\torch\nn\modules\module.py:1541 in _call_impl                   │
│                                                                                                                      │
│   1540 │   │   │   │   or _global_forward_hooks or _global_forward_pre_hooks):                                       │
│ ❱ 1541 │   │   │   return forward_call(*args, **kwargs)                                                              │
│   1542                                                                                                               │
│                                                                                                                      │
│ A:\Stable Diffusion\automatic\extensions-builtin\Lora\networks.py:398 in network_Conv2d_forward                      │
│                                                                                                                      │
│   397 │   network_apply_weights(self)                                                                                │
│ ❱ 398 │   return originals.Conv2d_forward(self, input)                                                               │
│   399                                                                                                                │
╰──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
TypeError: 'NoneType' object is not callable

Restarting stable diffusion fix it but this does happen often.

[vladmandic]

hard to say based on this traceback alone, i'd need more info, but from traceback it looks like its trying to apply lora that was never correctly loaded to start with.

[huangfanglong]

I noticed that this happens consistently when I used img2img to upscale (with new prompts specifically for details and quality only), then go back to txt2image to generate (using the original prompts with lora used), and then go back to img2img again to upscale that image using the details and quality prompts.