Unable to LoadNetwork, GPU HANG

[kirira]

Hi,

I installed OpenVino 2021.4.752 on Ubuntu 20.04, The used hardware is either an Intel NUC11TNHv5. I cloned ros_openvino_toolkit and switched to dev-ov2021.4 branch, All operations are normal in CPU mode. 

When I use the "GPU" mode, I have been waiting in the startup state without reporting errors

[ INFO ] --------parameters DUMP---------------------
[ INFO ] Pipeline: object
[ INFO ]        Inputs: ImageTopic, 
[ INFO ]        Outputs: RosTopic, 
[ INFO ]        Inferences: 
[ INFO ]                Name: ObjectDetection
[ INFO ]                Model: /opt/openvino_toolkit/models/public/mobilenet-ssd/FP16/mobilenet-ssd.xml
[ INFO ]                Engine: GPU
[ INFO ]                Label: to/be/set/xxx.labels
[ INFO ]                Batch: 1
[ INFO ]                Confidence_threshold: 0.5
[ INFO ]                Enable_roi_constraint: 1
[ INFO ]        Connections: 
[ INFO ]                ImageTopic->ObjectDetection
[ INFO ]                ObjectDetection->ImageWindow
[ INFO ]                ObjectDetection->RosTopic
[ INFO ]                ObjectDetection->RViz
[ INFO ] Common:
[ INFO ]        camera_topic: /camera/color/image_raw
[ INFO ]        custom_cpu_library: 
[ INFO ]        custom_cldnn_library: 
[ INFO ]        enable_performance_count: 0
[ INFO ] Parsing InputDvice: ImageTopic
[ INFO ] before Image Topic init
[ INFO ]  ... Adding one Input device: ImageTopic
[ INFO ] Adding Input Device into Pipeline: ImageTopic
[ INFO ] Adding connection into pipeline:[<-->ImageTopic]
[ INFO ] Parsing Output: RosTopic
[ INFO ]  ... Adding one Output: RosTopic
[ INFO ] Parsing Inference: ObjectDetection
[ DEBUG ] for test in createObjectDetection(), model_path =/opt/openvino_toolkit/models/public/mobilenet-ssd/FP16/mobilenet-ssd.xml
[ DEBUG ] model location:  /opt/openvino_toolkit/models/public/mobilenet-ssd/FP16/mobilenet-ssd.xml
[ DEBUG ] TESTING: in ObjectDetectionSSDModel
[ DEBUG ] for test in createObjectDetection(), Created SSDModel
[ DEBUG ] for test in createObjectDetection(), before modelInit()
[ INFO ] Loading network files: /opt/openvino_toolkit/models/public/mobilenet-ssd/FP16/mobilenet-ssd.xml
[ INFO ] Batch size is set to  1
[ INFO ] Checking INPUTs for model /opt/openvino_toolkit/models/public/mobilenet-ssd/FP16/mobilenet-ssd.xml
[ INFO ] Checking OUTPUTs for model /opt/openvino_toolkit/models/public/mobilenet-ssd/FP16/mobilenet-ssd.xml
[ INFO ] Checking Object Detection output ... Name=detection_out
[ INFO ] max proposal count is: 100
[ INFO ] ----Attributes for Model /opt/openvino_toolkit/models/public/mobilenet-ssd/FP16/mobilenet-ssd.xml----
[ INFO ] | model_name: /opt/openvino_toolkit/models/public/mobilenet-ssd/FP16/mobilenet-ssd.xml
[ INFO ] | max_proposal_count: 100
[ INFO ] | object_size: 7
[ INFO ] | input_height: 300
[ INFO ] | input_width: 300
[ INFO ] | input_names: 
[ INFO ] |    input-->data
[ INFO ] | output_names: 
[ INFO ] |    output-->detection_out
[ INFO ] --------------------------------
[ INFO ] This model is SSDNet-like, Layer Property updated!

Then execute `dmesg` to print GPU hang after a few seconds：

[ 3385.736754] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:1:85dcfffb, in pipeline_with_p [117144]
[ 3385.737778] i915 0000:00:02.0: [drm] Resetting rcs0 for stopped heartbeat on rcs0
[ 3385.737810] i915 0000:00:02.0: [drm] pipeline_with_p[117144] context reset due to GPU hang

Through simulation, I found that the program runs here: "core.LoadNetwork(model->getNetReader(), device);"

After installing openvino I have run ./install_NEO_OCL_driver.sh

To verify there is an Intel GPU, I did this:

 lspci | grep VGA 
00:02.0 VGA compatible controller: Intel Corporation Device 9a49 (rev 01)

How do I get the above example to run on the Intel GPU

Thanks for you help

[Peh_Intel · Accepted solution]

Hi Kirira,

Thanks for reaching out to us.

Can you check which GPU driver version you’re using? You can check using clinfo.

Please try to install this GPU driver version 21.29.20389 and see whether it can work.

Regards,

Peh

[Peh_Intel]

Hi Kirira,

Thank you for confirming your problem is resolved.

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.

Regards,

Peh