Intel® Distribution of OpenVINO™ Toolkit
Community assistance about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all aspects of computer vision-related on Intel® platforms.

Performance of NCS Sticks and OpenVino

idata
Employee
2,064 Views

I have been using the NCS 1 compute stick for some time now.

 

I installed OpenVino version 2018 4.420 on my older i5 Ivy Bride 1.7ghz 6GB DDR3

 

memory laptop with a single USB 3 port running windows 10.

 

The Demo Security Barrier Camera with no NCS sticks ran at 29ms ~ 34fps

 

The vehicle detection time was 11.8 ms

 

So I was looking forward to even better performance and much to my shock

 

with one NCS2 and one NCS1 I got ~10 fps.

 

Unbelievable, the CPU only version ran 300% faster.

 

Total time was 95.57ms ~ 10fps

 

The total vehicle detection time was 81.86 ms

 

With just the NCS 1 plugged in the performance was a bit worse 107.28ms ~ 9fps

 

Vehicle detection time was better 78.9ms ~ 12fps

 

With just the NCS 2 plugged in the performance was a bit worse 107.8ms ~ 9fps

 

Vehicle detection time was better 86.29ms ~ 11.7fps

 

Proving newer is not always better? I am totally perplexed by the results. I expected the NCS 2 to outperform the NCS 1 because of the additional shave cores intead the NCS 2 performed worse.

 

Movidius any explanations?

 

And why is my OLD i5 cpu performing 800% better with vehicle detection.

 

Is there something wrong with the demo code and MYRIAD?
0 Kudos
7 Replies
idata
Employee
1,639 Views
0 Kudos
idata
Employee
1,639 Views

So PINTO .what i am seeing is expected? A single 5 year old I5 CPU runs rings around TWO MOVIDIUS sticks? Have a hard time understanding that. The cpu is general purpose. The graph for movidus dropped down to fp16. What's the bottle neck. Or are the sticks that slow

0 Kudos
idata
Employee
1,639 Views

@chicagobob123

 

I do not know the internal implementation of OpenVINO, so I can not say exactly.

 

https://ark.intel.com/ja/products/65707/Intel-Core-i5-3317U-Processor-3M-Cache-up-to-2-60-GHz-

 

When using the CPU, I think that 2 Core + 4 Thread is utilized to the maximum.

 

My memory is ambiguous, OpenVINO seems to be using an API similar to NCSDK internally.

 

In other words, when using NCS, it always seems to be running in one thread.

 

I abandoned any implementation that already used NCS.

 

Sadly, I feel that development using NCS is barren until OpenVINO is compatible with ARM.
0 Kudos
idata
Employee
1,639 Views

ARM is coming soon. So do not toss the Pi just yet :smile:

0 Kudos
idata
Employee
1,639 Views

@chicagobob123

 

Yes, I know.

 

I am looking forward to. :)

 

https://github.com/opencv/dldt/issues/3
0 Kudos
idata
Employee
1,639 Views

Did you download opevino 5 for the pie?

0 Kudos
idata
Employee
1,639 Views

@chicagobob123

 

 

Did you download opevino 5 for the pie?

 

 

I downloaded it and installed it.

 

And it is operating normally.

 

It's the best!!

 

I plan to update my repository during today or tomorrow.

 

https://github.com/PINTO0309/MobileNet-SSD-RealSense.git

 

https://software.intel.com/en-us/articles/OpenVINO-Install-RaspberryPI

 

$ sudo apt update $ sudo apt upgrade $ wget https://drive.google.com/open?id=1rBl_3kU4gsx-x2NG2I5uIhvA3fPqm8uE -o l_openvino_toolkit_ie_p_2018.5.445.tgz $ tar -zxf l_openvino_toolkit_ie_p_2018.5.445.tgz $ rm l_openvino_toolkit_ie_p_2018.5.445.tgz
0 Kudos
Reply