Problems with Linux OpenVINO R5 GPU

[Kulecz__Walter]

First off it could easily be I've not done this last step in the GPU option setup:

"If you use 8th Generation Intel® Core™ processor, you will need to add:
   i915.alpha_support=1
   to the 4.14 kernel command line, in order to enable OpenCL functionality for this platform."

I have kernel 4.15.0 and I don't have a clue as to if an i7-4500U is 8th generation.  I'd rather not mess with the system boot up unless I know its absolutely necessary.

Second is it even worth trying to use the GPU on this chip?

Here is my sample code command, I believe its correct as its just the command I've used for CPU with CPU replaced by GPU:

~/inference_engine_samples/intel64/Release$ ./interactive_face_detection_demo -d GPU -m ~/intel/computer_vision_sdk/deployment_tools/intel_models/face-detection-retail-0004/FP32/face-detection-retail-0004.xml -d_ag GPU -m_ag ~/intel/computer_vision_sdk/deployment_tools/intel_models/age-gender-recognition-retail-0013/FP32/age-gender-recognition-retail-0013.xml -d_em GPU -m_em ~/intel/computer_vision_sdk/deployment_tools/intel_models/emotions-recognition-retail-0003/FP32/emotions-recognition-retail-0003.xml -d_hp GPU -m_hp ~/intel/computer_vision_sdk/deployment_tools/intel_models/head-pose-estimation-adas-0001/FP32/head-pose-estimation-adas-0001.xml
 

Here is the error:

InferenceEngine: 
    API version ............ 1.4
    Build .................. 19154
[ INFO ] Parsing input parameters
[ INFO ] Reading input
[ INFO ] Loading plugin GPU

    API version ............ 1.5
    Build .................. 19154
    Description ....... clDNNPlugin
[ INFO ] Loading network files for Face Detection
[ INFO ] Batch size is set to 1
[ INFO ] Checking Face Detection network inputs
[ INFO ] Checking Face Detection network outputs
[ INFO ] Loading Face Detection model to the GPU plugin
[ ERROR ] failed to create engine: clGetPlatformIDs error -1001
 

[nikos1 · Accepted solution]

Sorry Walter,

It seems that i7-4500U has an unsupported clDNN GPU:

https://ark.intel.com/content/www/us/en/ark/products/75460/intel-core-i7-4500u-processor-4m-cache-up-to-3-00-ghz.html

There was another thread in this forum explaining the issues with Intel HD Graphics 4400 or similar gen GPUs.

https://software.intel.com/comment/1931157

basically lacks OpenCL out-of-order execution   

cheers,

nikos

[Kulecz__Walter]

Thanks a lot, saves me much effort and wasted time, any insight as to how much a supported GPU improves on CPU only?

Initially for my purposes its looking like on i3 or better CPU outperforms the original NCS, and is still a bit better than the NCS2, but the NCS2 is about 2.5X faster than the original NCS. 

My next task is to test on the Raspberry Pi3B+ where I know an NCS will be required, and if I can get a 2.5X improvement in frame rate all this OpenVINO stuff will have been worth it.

[nikos1]

> any insight as to how much a supported GPU improves on CPU only?

It really depends on the CPU and GPU model.

Some GPUs like the GT4 in the Core i7-6770HQ for example are much faster than GT2 (like the (U)HD 630 ) ; all depends on how many EUs and GPU clock.

https://ark.intel.com/content/www/us/en/ark/products/93341/intel-core-i7-6770hq-processor-6m-cache-up-to-3-50-ghz.html

Comparing CPU to GPU again it depends how many CPU cores and what clock. If you compare to a dual core Atom CPU then GPU will be faster. There is also the power efficiency part.  It really depends on your application and fps vs. power requirements. 

Cheers,

nikos