OpenVINO cannot run on gpu device (stable diffusion Script)

[yeemoo]

I recently deleted my stable diffusion and reinstalled it with the OpenVINO toolkit for several times, this was the most recent problem I have encountered:

RuntimeError: Exception from src\inference\src\core.cpp:116:
[ GENERAL_ERROR ] Check 'false' failed at src\plugins\intel_gpu\src\plugin\program_builder.cpp:176:
[GPU] ProgramBuilder build failed!
bad allocation

I deleted my stable diffusion and reinstalled it using

git clone https://github.com/openvinotoolkit/stable-diffusion-webui.git
cd stable-diffusion-webui
webui-user.bat

I generated an image using a custom checkpoint as a reference point, it used around 6 minutes. Then I proceed to turn on "Accelerate with OpenVINO," selected "GPU" as my device, and "DPM++ 2M Karras" as my sampling method, others stayed the same. After around 1 minute, the error appeared.

I would like to know the solution to this problem.

Edit: I reloaded stable diffusion, using the exact settings from last time, the error now looks like this:

RuntimeError: Exception from src\inference\src\infer_request.cpp:231:
clWaitForEvents

I reloaded stable diffusion again, using almost the same settings from last time, the only difference is I changed the default width to 1024 and default height to 576, here's the error:

RuntimeError: Exception from src\inference\src\core.cpp:116:
[ GENERAL_ERROR ] [CL ext] Can not allocate nullptr for USM type.

I use Intel(R) UHD Graphics as my gpu.

[Megat_Intel]

Hi Yeemoo,

Thank you for reaching out to us.

For your information, I did not encounter any error and was able to generate an image using the configurations that you provided as shown below:

OS: Windows 11

Python: 3.10.6

GPU: Intel(R) UHD Graphics

 

This issue might be due to the checkpoint model itself. Can you please try to generate an image using the same configurations but with the default checkpoint model and see if it successfully generates any image?

On the other hand, could you also share the custom checkpoint model so that we can validate and investigate further?

Regards,

Megat

[yeemoo]

Hello Megat,

Here's the situation:

I tried to use the default checkpoint to generate images, the first one went well, but the second one encountered an error.

The image down bellow should explain what had happened, I do not understand what is clwaitforevents, please help me resolve this problem.

Sincerely, 

Yeemoo

[Megat_Intel · Accepted solution]

Hi Yeemoo,

I tried installing the same Python as yours (3.10.11), and reinstalled the Stable Diffusion. I was able to generate the images using 50 "sampling steps" multiple times without any errors.

From your error message, we suspect that the exception occurs when executing from the cached models. Can you try to untick the "Cache the compiled models on disk for faster model load in subsequent launches (Recommended)" button and see if this solves the issue of generating images multiple times?

Do let us know if you are still facing any issues, Thank you.

Regards,

Megat

[yeemoo]

Hello Megat,

I couldn't explain how excited I am right now.

I ticked of the cache compiled model option, and now i can generate multiple images without any errors. I even tried it on my custom checkpoint, it works just fine too.

However, I encountered a problem during the testing. I tried generate images in 1024 * 576, then I encountered another error, it appears to be a general error, the files down bellow are my screenshots about the whole situation.

Sincerely,

Yeemoo

[Megat_Intel]

Hi Yeemoo,

For your information, the error you received might be due to hardware limitations.

I tried to generate an image with the size of (1024x576) and it was successful. However, when I generated an image with the size of (1920x1080), I received the same error as yours and I noticed that my memory usage was at 99%-100%. The memory being fully occupied is likely the cause of the error.

 

Therefore, to avoid this error we suggest you generate images that are not too big and won't occupy your system memory completely. Hope this helps.

Regards,

Megat

[yeemoo]

Hello Megat,

Thank you for taking your time for helping me.

I had marked this topic as resolved, have a nice day!

Sincerely, 

Yeemoo

[Megat_Intel]

Hi Yeemoo,

Thank you for your question. This thread will no longer be monitored since this issue has been resolved. If you need any additional information from Intel, please submit a new question. Have a nice day!

Regards,

Megat