-
|
I need to generate a 1536* 1536 (or any big Hires) image, but it doesn't work. I am ready to do anything for this, even if it is generated for at least an hour. Are there extensions that make this possible? GPU: RX580 8GB Please help me. And another problem: (the first generation is at the bottom, there are no warnings) |
Beta Was this translation helpful? Give feedback.
Replies: 35 comments 6 replies
-
|
Well. There are few things you can try but don't expect much from your GPU. First of all i don't think using both |
Beta Was this translation helpful? Give feedback.
-
Please take a look at the screenshot, I think that I have configured something wrong, because I have lowered all the parameters that I saw. Another question: what does the And I also got a blue screen 2 times on which it was written about the lack of memory. (1 time 2 days ago, 1 time today, (when the SD was opened)) (the screen is blue like with an emoji ":(" ) |
Beta Was this translation helpful? Give feedback.
-
|
RX 580 is not good to generate such larger images as it upcasts 16-bit fp into 32-bit in hardware level. That means, it consumes more VRAM than Navi cards. However, you can try img2img upscaling that may use slightly less VRAM. |
Beta Was this translation helpful? Give feedback.
-
You deleted
I am not sure whats the best. You can lower Tile size even to 16 or 32 and then you can generate few images and check if you have visible seams. If it happens then you have to higher Tile overlap. Just experiment with it a little.
Never tried it
That shouldn't happen. Do you have swap memory set on your SSD drive ? I am pretty sure you will never be able to generate 2048x2048 image on your GPU but there is a little chance for 1536x1536. With Hires fix it shouldn't be a big deal. I was able to generate 1024x1024 images with SDXL model on RX570 4GB VRAM WITHOUT Hires fix and you have 4GB VRAM more than this. |
Beta Was this translation helpful? Give feedback.
-
Yes i agree but i noticed using |
Beta Was this translation helpful? Give feedback.
-
|
@Harbitos I will suggest you to try to generate in 768x768 and use img2img tab to upscale your image 2x with tweaked Tile size and Tile overlap settings. There is really nothing better you can do. And if you have problems with faces use Adetailer extension but it won't work on Edit: Also you can try to go into Settings/VAE and change encode and decode type for TAESD. Its way faster and doesn't change much. |
Beta Was this translation helpful? Give feedback.
-
Nothing helped me. 2. @TheFerumn yes, I deleted them and replaced them with the ones you said didn't help 3.
– it didn't help. 4.
– yes, it is set to "automatically". now I have 28GB, and at the time of departure it was 18GB, after closing the SD again 18GB. By the way, today I just generated an image (which my graphics card can generate - 1024*1280 without Hires.fix) and my computer memory froze again, the computer rebooted. 5.
– I did everything as you said by applying all the settings (generated 768*768 with Hires.fix x2). It generated 100%, but then the same error occurred (the console code is at the bottom). 6. I also asked a question in the issue: 7. If I can't solve the problem with high resolution, then most of all I want to at least solve problem with "non-cleaning graphics card", it really annoys me. |
Beta Was this translation helpful? Give feedback.
-
So you can't use Hires fix to 1536 resolution ? What's your settings ? Which upscaler you trying to use and which sampler ?
Check your GPU temperatures during generation. It totally shouldn't do it... Looks like it might be heating too much or SD agressively allocates memory for some reason. I actually never used Zluda on such old GPU. For RX570 i am only using DirectML and i don't have such issues. You can make experiment with another copy of Forge and check if its better on DirectML but it might be slower.
Its really old GPU and as i said above. Zluda might not work correctly with such tasks. Also memory leaks are nothing new in SD. |
Beta Was this translation helpful? Give feedback.
-
|
I was just testing on RX570 4GB VRAM. PonyXL model. Generated image 512x512 with DPM++2M sampler which is the best for such old GPU but you can also use Euler a. Upscaled it with Hires fix 3x with R-ESRGAN 4x+ up to 1536x1536 with Tile size 64 and default Tile overlap.
9 minutes for this task with DirectML and no issues. Actually generated it after previous try with lower Tile Size but it was way longer. 64 is pretty good. Just try to switch to DirectML and tell me if there is any difference. If it works it means you shouldn't use ZLUDA on your GPU |
Beta Was this translation helpful? Give feedback.
-
Bad and good news! And what's wrong, you said that some functions might not work because of (Of course, I already have a plan to have two venv folders, one for zluda, the other for directml, and change the arguments on the command line each time. I've even already made a comment in Question 1: which is generally better, directml or zluda? Question 2: some of my pictures seem to lack contrast, is it possible to fix or adjust it? The issue is not closed yet, we need to solve the problem with |
Beta Was this translation helpful? Give feedback.
-
With fp16 you won't be able to use inpainting and extensions which works the same way for example Adetailer. I am not really sure why its still a thing. Anyway you can just make another copy of webui-user.bat, rename it to webui-user-inpaint.bat and change commands from
Well as you can see ZLUDA is much faster but its intended to use with newer GPUs. This is why you had to download custom rocblas libs which might not work perfectly. So whats generally better ? Depends on what kind of GPU but generally ZLUDA is much better because its CUDA for AMD GPU and generally SD is meant to be used with CUDA GPUs so its optimized better for CUDA. |
Beta Was this translation helpful? Give feedback.
-
will the extensions work? |
Beta Was this translation helpful? Give feedback.
-
Generally everything suppose to work. I am not really sure why its a thing with inpainting. I am sure @lshqqytiger could tell more about it. All i know that there isn't any fix to it other than just changing commands which is not a big deal. If you really want to use Adetailer or Inpainting just use it with |
Beta Was this translation helpful? Give feedback.
-
While I'm testing, please answer the second question:
Maybe if I don't use zluda, I can delete the HIP SDK? |
Beta Was this translation helpful? Give feedback.
-
So its only when using DirectML ? Can you generate some example and show it here ?
If you are 100% sure you won't use it at all then you can delete it |
Beta Was this translation helpful? Give feedback.
-
Fact: Also, my computer does not lag using directml, but it lagged with zluda. Problem 1: You already told me to create a second webui-user, now it looks like this: — But when I start generating it, the console crashes without an error: If I download a non-forge version and start generating it, I will have other problems with SD "model is not loaded, trying slow metod..." there that I have not solved. I can't generate images in non-forge version, although I could have done it a long time ago. In addition, non-forge creates some kind of cache file in users/user, which then weighs 15-20 GB. (it works on zluda inpaint) Problem 2:
Problem 3 (probably): |
Beta Was this translation helpful? Give feedback.
-
Try to launch just with
I have no idea why you are having blue screens while using SD. Either your windows is cooked and you need fresh reinstall or your GPU might be heating too much.
Check here #13 . I already tried to fix this issue but i found nothing. Btw. if it happens to you too. Can you check if upscalling with img2img tab works for you ? |
Beta Was this translation helpful? Give feedback.
-
— not helped, SD also whites press any key to continue... does inpaint work for you?
img2img works fine, there are changes in the picture. The most important thing for me is to be able to do inpaint on 2048*2048 images. |
Beta Was this translation helpful? Give feedback.
-
So inpainting works for me with the commands i gave you. You didn't change anything in settings ? You need Btw. i never had this shown up in my console You was messing something with GPU Weights ? If this is not the reason then i don't really know why you have problems. You can also try to change checkpoint for inpainting SD1.5 to see if it works. You can do some simple inpaints on 1.5 model. |
Beta Was this translation helpful? Give feedback.
-
In short, I don't have time to answer you right now. I have everything as you asked in the settings. With the following arguments I managed to start inpaint generation! The change in the picture is really visible, but it says 100% that there is not enough memory. The picture has not been saved. I tried with --no-half-vae before, it didn't work, maybe with commands that didn't work now either. But I tried deleting the pip folder and installing 18. After that, I haven't tried with --no-half-vae yet. As for low warnong, I see SD in gpu-weights either 0 or 1024. If I changed them during inpaint, it didn't help. I have an sd 1.5 model, I had a bug with tensor and it couldn't generate. I used the dreamshaper_8 model. ran out of memory while running softmax in _get_attention_scores_no_kv_chunking, trying slower in place softmax instead |
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
(@lshqqytiger @TheFerumn) (I don't know how to notify you) This appears if I remove It seems that some kind of optimizer of my memory in SD does not work on directml, we need to fix it. I think that says it all: |
Beta Was this translation helpful? Give feedback.
-
|
What resolution are you trying to inpaint ? |
Beta Was this translation helpful? Give feedback.
-
(@lshqqytiger @TheFerumn) (I don't know how to notify you) The I set it to 0, first the console showed at the end |
Beta Was this translation helpful? Give feedback.
-
|
Use |
Beta Was this translation helpful? Give feedback.
-
I understand! My SD crashes due to the fact that I write lora in prompt, without lora everything works even 1280* 1024! (2048 *2048 I haven't tried yet). Question to you: do you have inpaint running on RX570 with loras? I completely reinstalled the SD, there was an error during the clean installation: I tried using ChatGPT, I don't think it helped. And I also had |
Beta Was this translation helpful? Give feedback.
-
Yes. Its working with LORAs. How many you tried ? Maybe its 1 specific LORA ? Maybe it has very high dimension and uses a lot of VRAM ? There might be a lot of reasons really.
Didn't help with what exactly ? You have so many different errors i am getting lost already... Its like everytime you write something you write another new error and i don't even know anymore what you trying to do. Sorry but its impossible to help you this way. You need to paste full logs of what you doing and your full generation settings otherwise its too big of a mess in this thread already... If you used different versions of python before and different python apps i feel like you might have some wierd leftovers outside your environment... Its stupid but it sometimes can happen and even deleting venv doesn't help so i suggest you to run this:
Just go into venv/scripts/activate and use this commands to delete everything again this way to be 100% sure you have everything clean. I feel like your PC is trashed too much... consider just reinstalling Windows if this won't help xD |
Beta Was this translation helpful? Give feedback.
-
|
btw. is your prompt is correct ? you written it manually or just clicked on Lora from the Lora tab ? Try to delete everything from prompts and leave only lora and use low resolution like 512x512 to see if it crashes too |
Beta Was this translation helpful? Give feedback.
-
|
I'm sorry that I was vague in my last message.
Nothing helped. I clicked lora in the tab lora. 512*512 and only using lora did not help.
This is the case with any lora. I draw attention to By the way, I was able to make inpaint 2048*2048 (only masked) without lora about 15-25 min! (time did not show me).
I do not know why this is happening, probably the SD repository is being updated. And what does the I didn't tell you about venv, it downloaded for about 25 minutes at a speed of 200kb/s. Although my internet speed is 300mbps/s And after reinstalling the SD, the sd1.5 model works for me! But it didn't work before.
I'll try it today. I don't know if this will help. I deleted the folder manually Why windows reinstallation? It hasn't even been a month since the last reinstall. I'm not the kind of programmer to go where I shouldn't. I'm not doing anything wrong with the PC. It is only after closing the old game and going to bed before shutting down that I press the keyboard shortcut Ctrl + shift + win + B — which restarts the graphics card driver. My computer is lagging after closing an old game, or I click this before shutting down before going to bed, because I have a PC for the whole day if I do shut down, it restarts.
If that's the case with any model! I've tried it. Generation with a resolution of 512* 512 and only lora in promt: |
Beta Was this translation helpful? Give feedback.
-
It shouldn't affect the bug you are having.
It updates your SD if there is any new update.
Yeah another post another new error xDDDD
Magic
Nevermind xD
This is new for me.... and very wierd. What game you are talking about ? |
Beta Was this translation helpful? Give feedback.
-
In the Plants vs zombies and Terraria games, OR some programs, but I don't remember which ones (it doesn't matter where I downloaded it from, it's the same everywhere). I think it has nothing to do with SD at all. And so think about it, if I have zluda Inpaint working with Lora, but directml does not, maybe something in directml is not fully configured? Did you tell me exactly everything that needs to be changed in the settings? I think the problems with the PC are not related to this. What if directml has noticed the ROCm libraries and is trying to use them? By the way, lora works in txt2img. |
Beta Was this translation helpful? Give feedback.
-
Well if you have some issues with GPU driver than it might be connected to the issues you had using ZLUDA. You shouldn't be forced to do this in Terraria or any other game. Its not that old.
with If you will follow instructions how to delete all packages with |
Beta Was this translation helpful? Give feedback.
-
I will try it now.
This was before I started using stable diffusion and before zluda. Maybe I should be lagging like this from the very beginning when I bought a video card, I didn't notice, it turns out to be probably brokaw. |
Beta Was this translation helpful? Give feedback.
-
Of course, I don't know much about it. I have registered all the commands separately. as a result, I pressed "y y y y y y" many times and eventually kind of deleted PIP. After all this, unfortunately, nothing has changed, it has not helped. By the way, venv download speed is back! Now 7-23mb/s I created an issue on this topic 46. I hope @lshqqytiger will help. |
Beta Was this translation helpful? Give feedback.
-
|
I am closing this discussion, the problem described at the very beginning has been solved here. |
Beta Was this translation helpful? Give feedback.
The solution to this problem:
Add to COMMANDLINE_ARGS:
--directml(by deleting thevenvfolder),--always-low-vramand--all-in-fp16 insteadInstead of--precision fulland--no-half(or add them if there are none).