Poor performance when running several models on HDDL device

[Morabito__Roberto]

Hi OpenVINO folks!

I have a system (UP Squared AI Edge) featuring a built-in Intel® Movidius™ Myriad™ X. 

I started playing with some of the demos in python. Once tried the "face recognition demo", I realized that when assigning the execution of all the models to the HDDL device, the performance are extremely poor -- basically the system is not able to process almost anything.

Is there any limitation on how many models HDDL can support? 

Thank you so much in advance.

Best Regards,
Roberto

[Sahira_Intel]

Hi Roberto,

About how many models did you run until you saw a drop in performance? 

Best Regards,
Sahira 

[Morabito__Roberto]

Hi Sahira,

Already with two models. The system get stuck basically. Is it possible that my VPU can't process more than a single model?

Thanks

Roberto

[Sahira_Intel]

Hi Roberto,

You should be able to run several models and still achieve good throughput on the VPU. When you're running the models, are you using multiple requests? 

Best Regards,

Sahira 

[Morabito__Roberto]

Hi Sahira,

Yes, I'm using multiple requests but it doesn't work. I mean, even using the "face recognition demo" released by Intel it does not work. Simply get stuck.
Apparently I'm not the only one facing this problem:
https://software.intel.com/en-us/forums/intel-distribution-of-openvino-toolkit/topic/842778#comment-1959932

Roberto