Ask Your Question

Movidius or a GPU would be better

asked 2018-02-06 14:56:08 -0500

Shivanshu gravatar image

please don't dislike,This is only meant for an advice

I am in need of some deep learning acclerators.Recentely i heard about intel's neural compute stick(movidius) and also heard that it supports opencv.Of the same price i found many nvidea gpu. My head is conflicting among these product. Shall i buy a GPU rather movidus or shall i buy movidus instead.I think gpu in a pc can do more than just deep learning,like games..:) From inside i feel to buy a gpu instead .but before any step i would like to take advice from expert or those who have experience about it.I want to judge on the basis of performance and experience

edit retag flag offensive close merge delete

2 answers

Sort by ยป oldest newest most voted

answered 2018-02-07 04:51:43 -0500

kbarni gravatar image

Movidius is good if you want to run a DNN in embedded environments (like the Raspberry Pi). It's like the AI processor in some smartphones (iPhone X, Huawei Mate 10), but packaged separately in a USB stick. As Tetragramm said, it won't train neural networks, but it's small with a low power consumption.

If you want to train AND use DNNs, you don't have space and power limitations (you are using a desktop computer) then a nVidia GPU will be much better. Buy the Movidius when your DNN is trained and tested, and you want to embed it (robot, car, intelligent camera, etc.)

edit flag offensive delete link more


Then shall i go with nvidea?.if i train a dnn then using with rasp. Pi will enhance its performanc? I also have a 2.7 ghz laptop 2nd gen processor .is that enoughf?

Shivanshu gravatar imageShivanshu ( 2018-02-07 11:26:27 -0500 )edit

Are you joking, right???

No, you can't add a NVidia video card (it looks like this) to a Raspberry Pi. You can't add it to your laptop neither. You need a desktop computer for that.

And forget about training a neural network on the Raspberry Pi. You can do it on a laptop, but it will still take 10-20x more time than the GPU (a 5th gen. core i7 CPU compared to the nVidia 1060 GPU).

kbarni gravatar imagekbarni ( 2018-02-08 05:16:13 -0500 )edit

no i meant to say if i train dnn with gpu and implement it on rasp.pi with movidus .this is what iwas saying about .I this possible???

Shivanshu gravatar imageShivanshu ( 2018-02-08 14:00:52 -0500 )edit

Yes, it's possible.

kbarni gravatar imagekbarni ( 2018-02-09 11:04:19 -0500 )edit

Thank you @karbani

Shivanshu gravatar imageShivanshu ( 2018-02-10 06:18:57 -0500 )edit

answered 2018-02-06 22:14:26 -0500

Tetragramm gravatar image

This is really not the place, but you should note that the Movidius doesn't train neural networks, just run already trained ones.

edit flag offensive delete link more


This is a forum .People with questions and knowledge come here together to share whatever they know, if they can.Stop saying like this is not a place or not a good question bro!.This is place of expert like you and I gave a problem which is not off-topic.Then it dosent meant to say things like this..

Shivanshu gravatar imageShivanshu ( 2018-02-10 06:24:38 -0500 )edit

@Shivanshu, I think it was just a game of words. Perhaps, @Tetragramm wanted to say that OpenCV doesn't provide tools to train deep learning networks and/or deploy them on NVIDIA GPUs or Intel Movidius. Currently you need to use specific software. So it's hard for OpenCV community to recommend you one of them because non of them is supported by OpenCV for training (GPU) or deployment (NCS or GPU) of neural networks.

dkurt gravatar imagedkurt ( 2018-02-10 06:44:24 -0500 )edit

Do opencv don't provide GPU support at all ????

Shivanshu gravatar imageShivanshu ( 2018-04-01 23:49:11 -0500 )edit

OpenCV has plenty of GPU support, but OpenCV is not a deep learning library. The DNN module(s) do not support the GPU, nor do they support training a network, nor do they support anything like the Movidius. Hence, not really the place.

There are very well made libraries such as tensor flow or theano, and excellent wrappers like Keras to help train and use neural networks. There is little reason for OpenCV to compete with them.

Tetragramm gravatar imageTetragramm ( 2018-04-02 16:44:40 -0500 )edit

Actually, OpenCV's deep learning module supports GPU devices using OpenCL. You may enable DNN_TARGET_OPENCL as a preferable target for computations. Also there is a PR which enables OpenCL target (clDNN) for Intel's Inference Engine backend.

dkurt gravatar imagedkurt ( 2018-04-03 02:42:17 -0500 )edit

Question Tools

1 follower


Asked: 2018-02-06 14:56:08 -0500

Seen: 4,323 times

Last updated: Feb 07 '18