RuntimeError: Failed to create plugin for device GPU with Ubuntu 20

[Lucho]

Hi,

Working with EII 3.0, Intel(R) Core(TM) i7-8550U CPU and Ubuntu 20.04.

I have already published a message regarding this problem:

RuntimeError: Failed to create plugin for device GPU 

Now EII 3.0 has support for ubuntu 20.04 as It is named in the requeriments.

The problem with GPU persist in this EII version and I'm not able to execute PySafetyGear customUdf Analytics with GPU device.

Any help will be appreciated.

Regards,

Luciano

[Sahira_Intel]

Hi Luciano,

I am looking into this for you but have a few questions.

Are you running into the same runtime error as before? 

Were you able to successfully run PySafetyGear customUdf Analytics on Ubuntu 18.04?

Sincerely,

Sahira

[Lucho]

Hi @Sahira_Intel ,

Yes, I were able to run it on Ubuntu 18 but I have the same error as before in Ubuntu 20 with EII 3.0.

Regards,

Luciano

[Lucho]

Sorry, I were able to run it on ubuntu 18 with EII 2.6.2. With EII 3.0 on ubuntu 18 I am not even able to install EII due to this error:

Error loading Python lib '/tmp/_MEIj6cJif/libpython3.8.so.1.0': dlopen: /lib/x86_64-linux-gnu/libm.so.6: version `GLIBC_2.29' not found (required by /tmp/_MEIj6cJif/libpython3.8.so.1.0)

[JesusE_Intel]

Hi Luciano,

I believe this may be a documentation error and Edge Insights for Industrial v3.0 requires Ubuntu 20.04. When downloading the EII v3.0 package on Edge Software Hub, Ubuntu 20.04 is listed as the OS. Could you try installing EII v3.0 on Ubuntu 20.04 and let me know if you have any issues?

In the meantime, we will check with the development team if this is a documentation issue or a bug with the installer.

Regards,

Jesus

[Lucho]

Hi @JesusE_Intel , 

Yes, We have installed EII 3.0 on Ubuntu 20 without any problems and as it is mentioned in the title, we are having problems with GPU on ubuntu 20.

Regards,

Luciano

[JesusE_Intel]

Thanks, let me setup a system to reproduce. In the meantime, could you share your kernel and GPU driver version?

[JesusE_Intel]

Hi Luciano,

Apologies for the delay, I was able to resolve [CLDNN ERROR]. clGetPlatformIDs error -1001 by adding user: root to the docker-compose.yml file in VideoIngestion and VideoAnalytics. Please give that a try and see if that works for you too.

I've run into a new issue but it may be specific to my environment. Let me know if you run into any new issues.

Reference: https://open-edge-insights.github.io/pages/running.html#to-run-on-intel-r-processor-graphics-gpu-igpu

Regards,

Jesus

[Lucho]

Hi @JesusE_Intel ,

We made that work around as I mentioned in the previous issue and it works, but we were trying to get a solution using eii_user. Thanks for the reply.

Regards,

Luciano

[JesusE_Intel]

Hi Luciano,

I believe this is a requirement as docker requires elevated permissions to access the hardware outside of the container. However, let me check with the development team to confirm this is the case.

Regards,

Jesus

[Lucho · Accepted solution]

Hi @JesusE_Intel ,

we were able to make it work using:

sudo chown root:video /dev/dri/renderD128

on the host.

Regards,

Luciano

[JesusE_Intel]

I have confirmed with the development team that inferencing on GPU need to run as root to give the necessary permission to access the GPU resources. The EII user will only be created during the runtime and being destroyed when all the container being killed. Glad you were able to get it to work.

If you need any additional information, please submit a new question as this thread will no longer be monitored.