When I run inference using device « MULTI:CPU,MYRIAD » (on Python or wih benchmark_app) I get the same inference time than with « CPU » only. I get the same problem with two Myriad devices : « MULTI:MYRIAD1,MYRIAD2 » I get the same results than with one Myriad : « MYRIAD ».
Do you know how to resolve this pb ? Thanks for your help :)
Enumerate your available devices using the OpenVINO Hello Query Device Sample to ensure they actually exist and are usable.
Please note that certain device combinations would result in performance caveats, as these devices share the power, bandwidth, and other resources. For example, in combining CPU and GPU, it is recommended to enable the GPU throttling (which saves another CPU thread for the CPU inference)
The reason why you've got the same inferencing time might be because the program falls to the available device which ends up being the same. For example: for CPU + MYRIAD, the program might unable to detect the presence of MYRIAD (or it's actually unavailable), hence it is executed using CPU.
This might help.