The question about the detection speed of nvidia 1660 6G and 1080 Ti 6G

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )

http://pjreddie.com/darknet/

Other

21.8k stars 7.97k forks source link

Open jwnam-ct opened 4 years ago

jwnam-ct commented 4 years ago

Dear all.

My name is Jung Woo, Nam.

We are develop of security software in South Korea.

Now we are under developing video security software with YOLO v3(AlexeyAB/darknet).

Our plan is that we want to integrate your YOLO v3(AlexeyAB/darknet) engine into our security software applications.

I have some questions.

I was test the YOLO v3(AlexeyAB/darknet) with nvidia 1080Ti 6GB and nvidia 1600 6GB. But the detection speed of nvidia 1660 6GB was low less than nvidia 1080Ti 6GB. I don't know why nvidia 1660 has better specification performance but the detection speed was low.

If you have some questions, please let me know. Have a good day.

AlexeyAB commented 4 years ago

Did you compile with GPU=1 CUDNN=1 CUDNN_HALF=1 OPENCV=1 in both cases?

GeForce GTX 1080 Ti - 11 TFlops-SP
GeForce GTX 1660 - 5 TFlops-SP (it doesn't have Tensor Cores https://www.tomshardware.com/reviews/nvidia-geforce-gtx-1660-turing-tu116,6027.html

Because GeForce GTX 1660 doesn't have Tensor Cores, then GeForce GTX 1080 Ti should be faster.

jwnam-ct commented 4 years ago

Dear sir.

Thank for your cooperation.

Yes, I compiled with GPU=1 CUDNN=1 CUDNN_HALF=1 OPENCV=1.

And I understood that.

I wishes for a happy new year. And have a good days.

jwnam-ct commented 4 years ago

Dear Sir.

I tested the nvidia GTX 1060 6GB(cuda cores : 1280) and nvidia GTX 1660 6GB(cuda cores : 1408).

However the GTX 1660 had high performace but a low yolo detection speed.

The GTX 1060 recognition faster than GTX 1660.

I am asking why this works.

If you have some question, please let me know.

Have a good days.

AlexeyAB commented 4 years ago