GPU works with 2021.3 on Ubuntu20.04 but not Ubuntu 22.04 and 2024.3

[wb666greene]

I have a couple of N4020 Celeron with Mesa UHD 600 graphics (GLK2) systems running Ubuntu 20.04 and OpenVINO 2021.3 that run Darknet Yolo4 on the iGPU.  I updated one of them to Ubuntu 22.04 and pip installed openvino 2024.3 in a VENV virtual environment to run Ultralytics Yolo8 on the iGPU.  But running code that works fine on other systems I get this error:

[INFO] Using OpenVINO: 2024.2.0-15519-5c0f38f83f6-releases/2024/2
Available devices:
   CPU: Intel(R) Celeron(R) N4020 CPU @ 1.10GHz
   GPU: Intel(R) UHD Graphics 600 [0x3185] (iGPU)
Traceback (most recent call last):
  File "/home/ai/LearnTestOV/yolo8ov.py", line 45, in <module>
    det_compiled_model = core.compile_model(det_ov_model, "GPU", ov_config)
  File "/home/ai/xyz/lib/python3.10/site-packages/openvino/runtime/ie_api.py", line 543, in compile_model
    super().compile_model(model, device_name, {} if config is None else config),
RuntimeError: Exception from src/inference/src/cpp/core.cpp:104:
Exception from src/inference/src/dev/plugin.cpp:54:
Check 'false' failed at src/plugins/intel_gpu/src/plugin/program_builder.cpp:185:
[GPU] ProgramBuilder build failed!
Exception from src/plugins/intel_gpu/src/runtime/ocl/ocl_stream.cpp:429:
[GPU] clWaitForEvents, error code: -14

Note: I saved this error message from 2024.2, but upgrading to 2024.3 didn't change anything, so I'm posting this question now since installing all these "neo" driver packages and rebooting didn't change anything either:

https://github.com/intel/compute-runtime/releases/tag/24.31.30508.7 

Edit:  Should mention that running kernel:

6.8.0-40-generic #40~22.04.3-Ubuntu SMP PREEMPT_DYNAMIC Tue Jul 30 17:30:19 UTC 2 x86_64 x86_64 x86_64 GNU/Linux

[Aznie_Intel]

Hi Wb666greene,

Thanks for reaching out. What version of your OpenCL drivers are you using? The validated version with the latest OpenVINO is 22.43 for Ubuntu 22.04.

You can find out this using 'clinfo'

Regards,

Aznie

[wb666greene]

clinfo dumps out a lot, I'll upload a clinfo.txt file of what it says.

Not much of it means anything to me, but it does say: "Platform Version  OpenCL 3.0"

I noticed this is a "Gemini-Lake" system, which on the page I got the "latest" drivers from, says Quality: "Maintenance" and there are "--" for "level zero" in the table.

I may be out of luck in terms of running yolo8 on its iGPU,  my project is here:

https://github.com/wb666greene/AI-Person-Detector-with-YOLO-verification-Version-2 

I can run it with two Coral TPUs, but at $60 for a USB TPU it basically kills the economy of using this system as a host for four or fewer cameras.  The previous version of my system runs on this iGPU using Ubuntu 20.04 and OpenVINO 2021.3 and yolo4, but yolo8 is demonstrably better than yolo4 for my use case. 

[Aznie_Intel]

Hi Wb666greene,

You can try running the following commands on a Linux* terminal. If the output is empty, then the compute-runtime driver is not present in the system. 

   $ export gfx_version=$(dpkg-query --showformat='${Version}' --show intel-opencl-icd)

   $ echo $gfx_version

Regards,

Aznie

[wb666greene]

Thanks, here are the results:

ai@hb2:~$ export gfx_version=$(dpkg-query --showformat='${Version}' --show intel-opencl-icd)
ai@hb2:~$ echo $gfx_version
24.31.30508.7

I'd for there to be a solution for this.

[Aznie_Intel]

Hi Wb666greene,

We are checking this with the engineering team and will get back to you once the information is available.

Regards,

Aznie

[Witold_Intel]

Hello Wally,

I see that you're using an older version of the compute runtime driver. Do you have an option to update it to the latest one? https://github.com/intel/compute-runtime/releases/tag/24.35.30872.22

[wb666greene]

Thanks for your response,  I'd really like to get this to work if possible.

Clearly the driver packaged with Ubuntu 22.04 didn't work for this system, although it did for the other six iGPU systems I've tried, although this is the "lowest spec" system I've tried yet.  I had installed the 24.31 version following a link in some on-line doc I'd found Googling the error message before posting the original message.  I redid it with 24.35 drivers and it still looks like the same error:

~$ echo $gfx_version
24.35.30872.22

image 1/1 /home/ai/LearnTestOV/AI/TestDetection.jpg: 512x640 1 person, 1 snowboard, 2105.5ms
Speed: 5.4ms preprocess, 2105.5ms inference, 710.6ms postprocess per image at shape (1, 3, 512, 640)


[INFO] Using OpenVINO: 2024.3.0-16041-1e3b88e4e3f-releases/2024/3
Available devices:
   CPU: Intel(R) Celeron(R) N4020 CPU @ 1.10GHz
   GPU: Intel(R) UHD Graphics 600 (iGPU)
Traceback (most recent call last):
  File "/home/ai/LearnTestOV/yolo8ov.py", line 46, in <module>
    det_compiled_model = core.compile_model(det_ov_model, "GPU", ov_config)
  File "/home/ai/y8ovv/lib/python3.10/site-packages/openvino/runtime/ie_api.py", line 543, in compile_model
    super().compile_model(model, device_name, {} if config is None else config),
RuntimeError: Exception from src/inference/src/cpp/core.cpp:104:
Exception from src/inference/src/dev/plugin.cpp:53:
Check 'false' failed at src/plugins/intel_gpu/src/plugin/program_builder.cpp:185:
[GPU] ProgramBuilder build failed!
Exception from src/plugins/intel_gpu/src/runtime/ocl/ocl_stream.cpp:429:
[GPU] clWaitForEvents, error code: -14

Frustrating,  that I could use the iGPU on this system with OpenVINO 2021.3 for Darknet YOLO4 on Ubuntu 20.04 but not the much improved Ultralytics YOLO8 with and 2024.3

I guess I can try 2024.4 as I just got a notification that is has been released, but if it is a driver problem I doubt it'd help.  I'll do it eventually as I need to verify that 2024.4 doesn't break anything in my code, going from 2024.2 to 2024.3 was fine.   

It seems like it is something in the cpp code that python calls, in the past I've had issues with cmake versions, this system has cmake version 3.22.1 if it might matter.

[Witold_Intel]

Thank you for your feedback. Would you still like to attempt with 2024.4 or should I consult this case with developers?

[wb666greene]

I just upgraded this system to 2024.4, made no difference.  If you think the "Gemini Lake" iGPU should work I'd like a solution if there is one.

[Witold_Intel]

Thank you for your reply. I opened an issue with the developers in order to resolve this case fully. As soon as I learn more, I will get back to you.