I am using NCS2 from last month for my segmentation code with keras. I have finally built successfully with my own Unet model in openvino.
Now the problem is,
- Keras code in python platform takes 0.75 seconds
- Openvino with MYRIAD(NCS2) takes 0.53 Seconds
- Whereas Openvino with CPU(Intel Xeon) takes only 0.08 Seconds
Is this how MYRIAD performs usually? Or am I missing any additional Plugins? How CPU with Xeon processor performs with 0.08 seconds and NCS2 takes 0.53 seconds. Its huge difference. Isn't it?
Note : I am using openvino_2019.1.148 version only.
Hi Suresh M,
How much Xeon cores OpenVino occupies to reach 80ms?
And what is the base frequency for it?
I think that your Xeon may have some another tasks during inference in real world, right?
It's up to you to decide which device to use for inference in real conditions.
Actually such performance comparison (Myriad and Xeon) should be recalculated to perWatt consuming (and perDollar cost :)). Right?
Dear om77 yes, thanks for your insightful input.
Dear Muthusamy, Suresh
MYRIAD performance cannot be reasonably compared to an Intel Xeon CPU right ? There is definitely a power/performance trade-off which you need to consider before you select devices for inference. But OpenVino does offer the benchmark_app which you can use to run various experiments. Please give it a try and report your findings here.
Thanks for using OpenVino !