Fixed- Torch is not able to use GPU AMD Stable Diffusion --skip-torch-cuda-test (AUTOMATIC1111) Bug 

Excellent tutorial! Very good!!! It worked here, and I can use my 4GB AMD GPU again. Thank you, I appreciate you sharing this knowledge. I left a like and you gained another subscriber! Congratulations on the channel 🙌🙏

Hello, I'm from Brazil, I'm infinitely grateful for your tutorial, not even on the official AMD website do they teach this, thank you very much for the video, all the best to you!

I have taken every step to install automatic 1111 on AMD VGA according to the video on this channel, but when I try to execute the webui-user.bat file using administrator still "torch is not able to use gpu" when I try to skip the gpu check, it can open the automatic window 1111, but still when rendering the image midway still and there is word 'error' torch is not able to use gpu I don't know what's wrong, maybe installing a new torch is not accompanied by uninstalling the old torch... but I can't fix it either. 😪

Thank you! I updated SD and it stopped working and this fixed my issue! My AMD isn't as fast as Nvidia at this stuff, but it's still way faster than using my CPU.

Ty i struggled for like 3h, gave up then some day later found your video and its working

Absolute legend! I've watched about 20 different videos on this issue and yours worked! Thank you!!!

Huge ups, brother! Thank you so much! Also, I love the accent. Have a great day!

Hi, After I edit the requiremenets_version.txt. and run the install requiremenets, I get this error: ERROR: Could not find a version that satisfies the requirement torch-directml (from versions: none) ERROR: No matching distribution found for torch-directml

great video. I had some issues and want to add what fixed them for me in case people still have issues. After everything in this video I also had to instal Visual Studio Community and delete the vent folder and let it recreate it on first launch

You Sir, are a godsend. Fixed the issue after 3 hours of different shitty videos explaining it poorly. Thank you so much.

Fixed my issue, thanks for the video!

Eres mi nuevo Dios, trabajé perfectamente. Pasé como una semana buscando alguna solución, hasta que encontré tu video. Eres el mejor, muchas gracias amigo. rx 580 8gn nitro

Thanks !! Worked like a charm !

hello i was following your video and realized that for some reason i do not have webui.user.bat on the file in order to edit it. what should i do? And thanks in advance!

Thank you so much! after so many videos and blog posts this worked for me, running RX 7800 xt, however still not able to run olive

THANK YOU. You're a genius and an absolute credit to this community. Can't thank you enough for your amazing help, wish you the best!

THANK YOU SO MUCH. I was searching for a solution but nothing worked. now everything is working really good. I still had some error, but changing the Arguments to set COMMANDLINE_ARGS=--use-directml --medvram --precision full --no-half --no-half-vae --opt-split-attention-v1 --opt-sub-quad-attention --disable-nan-check worked for my 6700 xt

For people that don't know how to open a given folder in CMD: click the big bar in top of your folder, then write there "cmd". Then the folder will open in CMD by itself.🎉

thank you very much! at first i thought my system is trolling me but i just had a typo in the requirements file and now its working :D

Will you do a video when the bug is fixed? I can't get it to work still, the "venv\Scripts\activate" line does not work and I don't know what I am doing wrong.

I did this and it seems fixed but in the terminal it says it's running with torch 2.0.0+cpu. Does this mean my GPU is not being used?

Sir u dont khow how much u have helped me ive been running errors for 4 day and u solved every problem with the codes

Thank you very much, the problem still exists and after reinstalling repository and making all steps in this video i finally can use Stable Diffusion, after six hours of trying to solve the problem!

Thank you very much for this tutorial

Thank you for your video man ! Nice explanations and thanks to this videos comments all my problems were fixed, I just don't know if removing the "--onnx" command will change something but now my CPU can breath !

I added exactly the same command lines on the proper files and the error still persists: Traceback (most recent call last): File "G:\StableDiff\stable-diffusion-webui\launch.py", line 48, in main() File "G:\StableDiff\stable-diffusion-webui\launch.py", line 39, in main prepare_environment() File "G:\StableDiff\stable-diffusion-webui\modules\launch_utils.py", line 384, in prepare_environment raise RuntimeError( RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check

hey thank you very much. took forever till I realized my graphics card is the problem. After hours this video finally gave me a solution.

How do i open an Administration Command Prompt- webui-iser.bat?? You skipt it in the video and now I can't continue, please help!!

I spent hours looking for a solution, thank you!

If you get the error: "AttributeError: module 'torch' has no attribute 'dml'"; Return to the commandline virtual environment and run the following: pip install torch-directml

yes, this worked, ty so much. using amd rx6600

I used the first video to reinstall SD. Then I used this video to fix the issue. Happy to report that after 3 months, Stable Diffusion is using my AMD gpu again!! Thanks for the straight forward video!!

thank you gatão

I now run into a new problem "AttributeError: module 'modules.devices' has no attribute 'inference_context'". Any ideas?

This is a godsent, thank you! I followed the guide to the dot and it now completely uses the 7800xt instead of CPU. The only issue I have now that I need to use the administrative cmd in order to open SD, else I'll get a fatal: detected dubious ownership in repository error.

dankie man dis nie julle mal die selfte probleem as wat ek het nie maar dit het my n paar goeie tips gegee

may you be blessed with eternal health :) why make it complicated when you can always show it as easy as you do, thank you :)

I get the FileNotFoundError: "Can't run without a checkpoint.", even though i placed the models into the models/stable-diffusion folder. I would be really happy about a solution!

Thanks for your help, it's have fixed my issue, Now it works much faster with my 7900GRE but i had to remove the cmd --onnx that create me an other issue

This video saved me so much time thx

I am now running A1111 DIRECTML version using ZLUDA (--use ZLUDA) Setup is hard, but is way more flexible as it uses ROCm as "cuda" cores. Only about 20% slower than using ROCm in Linux

Hi! The error "FileNotFoundError: No checkpoints found. When searching for checkpoints, looked at: - file G:\SDamd\model.ckpt - directory G:\SDamd\models\Stable-diffusionCan't run without a checkpoint. Find and place a" .ckpt or .safetensors file into any of those locations" how to fix this? There are models in the Stable-diffusion folder (openjourney_V4.ckpt and ghostmix_v12.safetensors) but he doesn’t see them (

EDIT: I think I fixed it? See below... Can anyone help please? Performed steps in original video and in this video. My issue is that is seems like my install is ignoring any and all changes that I make to my webui-user.bat. I'm running Windows 11 and have a Ryzen 7 7840 with an 780M GPU. Ultimately, when I run webui.bat, I still get "RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check" - Its maddening I've even tried adding these arguments to my webui-user.bat and it still gives me the same mindless error about "not able to use GPU" X_X: --use-cpu all --precision full --no-half --skip-torch-cuda-test Of course, I first tried only "--use-directml" but it doesnt do anything. Any help would be appreciated. Thank you I eventually updated my webui-user.bat to the following args: set COMMANDLINE_ARGS=--use-directml --opt-sub-quad-attention --no-half --disable-nan-check --autolaunch Then I STILL couldnt run webui.bat, got the same error, so something is definitely wrong with the script somewhere... HOWEVER, I then went into command prompt into my stable diffusion git-cloned directory and from command prompt I simply ran "webui-user.bat" and I waited forever for it to download and install things, and I walked away from my PC and like 10 minutes later, I came back to my PC and the webui was open! I then selected stable diffusion 1.5 from the dropbox, and typed in "a bear" and I saw my GPU being utilized 100% and in under 1 minute I had an Ai-made 'bear' image! Good times! :) In other words, once you have everything installed the first time, DONT use the "webui.bat" to open the Web UI....use the webui-user.bat file, and if you followed the other instructions in this video here, it should open fine after some time. Good luck to everyone else, hope this maybe helps.

Thank you for your help.

It worked ty so much!

There's no file named *requirements.txt* in the *Scripts* directory

For those saying it doesn't work, make sure you git clone the specific git repo linked in this video. Making the changes on the original A1111 repo will not work.

Works nicely with my 6700 XT :) just need to spend time fine-tuning with different settings etc.

THIS UNIRONICALLY WORKED BRUH!!!!!! YOOOOOOOOOOOO

I have problem AttributeError: module 'torch' has no attribute 'dml', how fix that?

Thank you.

Funcionou perfeitamente meus parabéns nota 10

You fuckin legend. Edit: worked with my 6700xt

Hi, I did everything as in the video and I get a new error "ModuleNotFoundError: No module named 'pillow_avif', I installed the library with "pip install diskcache" and "pip install Pillow-avif-plugin" and the installation was successful, but the error persisted. Can you tell me how to fix it?

ERROR: Could not find a version that satisfies the requirement torch-directml (from versions: none)

You really saved my day.

en el archivo webui-user.bat quite el --onnx y funciono ...gracias mil!

Hey i love your video and uncle ho as well

you are a gOD! fixed my problem

I has to come back to thank you

Hi. I followed every step 3 times, but it didn't help. I have an AMD processor with an integrated GPU, however I also have an RTX series GPU. Maybe stable diffusion doesn't see my external GPU and only sees CPU's GPU?

thanks to you my pc goes brrrrrr for 4 weeks now haha

me ayudo, lo estoy haciendo desde una amd rx5500 con 8 gb de ram y ahora si dejo de usar la cpu para correr con la gpu y es una maravilla :D, dejame invitarte un cafe :D

Hablo español, pero lo explico de manera tan grafica que hice paso a paso lo que dijo y funciono. Muchas gracias

Why do you say it is a temporary solution? The problem continues to affect operation in some way

THANK YOU

Work for me doing 2 changes: 1) I didn't add --onnx to webui-user.bat 2) I didn't install pip install httpx==0.24.1 (I keep with 0.24.0) Man, I tried for 7 hours to install SD without success. Your video saved me. Thank you very much for the contribution, it has been very helpful. Thank youuu!!!!!!!

I followed all the steps but when i run the bat file it still shows me the same error

Tutorial was almost perfect! You didn’t really explain how to get the the command arg you just appeared there..

Thank you so much!!!!!!

all worked except requirements.txt why would that fail? i typed pip install -r requirements.txt Also note there is already a requirements.txt file in SD folder so do I have to do this step? Does the command ONLY add something to the existing requirements.txt? if so what does it add?

I followed the steps as is but I have two problems, if I put --onnx in the bat it tells me that it does not recognize the --onnx command, if I put COMMANDLINE_ARGS=--use-directml --medvram --precision full --no-half --no-half-vae --opt-split-attention-v1 --opt-sub-quad-attention --disable-nan-check as I found in the comments it shows that module torch has no attribute dml

2:08 How to change directory in prompt ? I don't quite understand

I have an error : ERROR: Could not open requirements file: [Errno 2] No such file or directory: 'requirements.txt'

If you're using Stability Matrix like me, literally just update the SD package. I tried everything but that for hours but that alone works

So am I using administrator command prompt instead of anaconda prompt?

I followed the directions and I get an error ImportError: DLL load failed while importing onnxruntime_pybind11_state: A dynamic link library (DLL) initialization routine failed. Press any key to continue . . . How do i fix it?

can you fix it in ubuntu 22.04?

every command worked up until 3:29 which is telling me Torch is still not able to use GPU, any tips?

friend you can help me with another problem, I do not recognize the safetensor models, only those compatible with onnx, I already saved the models in their corresponding files and still nothing, edit the .bat file by placing this command and nothing. SATENSORS_FAST_GPU = 1

thank you! its woorkkk!!!

Excellent tutorial! Worked, THX !!!!

bro thank you u love u

you're the GOAT

ERROR: Could not open requirements file: [Errno 2] No such file or directory: 'requirements.txt' help

It somehow works now, but only with the argument --use-directml. It is very slow, i am using a 7900 xtx and have under 2it/s. If i add the argument --onnx, then i get the message "no module 'xformers'. Processing without..." and the error "launch.py: error: unrecognized arguments: --onnx" i did the installation step by step. What went wrong? Edit: After i started it with the argument --xformers the error is gone, but now i can no longer start it without that argument. The speed is still under 2 it/s. why doesn't it recognize the --onnx argument?

i tried every command you said and all are installed succefully but it still showing " torch not able to use gpu" can u help

LEGEND !!!!!!!

Hey I have an issue, It only let's me use the base model not sdxl

I followed this video about fix, and still getting error: launch.py: error: unrecognized arguments: --onnx --onnx is not an option in launch.py At least for me :( How to get onnx tap?

I also followed all the steps but when i run the bat file it still shows me the same error

three months later and still not fixed.

× Getting requirements to build wheel did not run successfully. │ exit code: 1 ╰─> See above for output. note: This error originates from a subprocess, and is likely not a problem with pip. can you help me?

why webui user not run as administrator

bro.. i can't do the process in [venv/Scripts/activate]... what can i do??

pls put the commands in the description...

i did all of them but finaly write Torch is not able to use GPU .... damn to this chance .my graphic is amd what should i do???????

May i know how im gonna open that cmd?

thank you. worked flawlessly, hopefully the github repo is updated soon.

Serves for 6800xt?