I have bought two graphics card for custom object detection using yolo-v3.

1) GTX 1060 2) GTX 1660 According to the specification GTX 1660 graphics card is better than GTX 1060. But while testing yolo-v3 model GTX 1060 is 5 times faster than GTX 1660. To detect object from 1 small image in GTX 1060, it takes 35 millisecond To detect object from 1 small image in GTX 1660, it takes 140 millisecond I have used CUDA10.0 and CUDNN 7.4 for yolo_cpp_dll. According to specification GTX 1660 should be faster than GTX 1060. Am i wrong in somewhere please comment. Regrads Israfil

  • The 1660 completely matches or beats the 1060 in all of the raw specs (cores, frequency, TFLOPS, etc.) So I would expect the 1660 to be slightly faster too! The 1060 only supports CUDA 6.1, while the 1660 supports CUDA 7.5 so building the same program for both certainly results in different CUDA code. Any chance there is a regression or difference in the CUDA versions that you haven't noticed? – Romen Sep 10 at 16:31
  • Also are these tests done with one GPU installed at a time on the same system? Or both installed in separate slots on the same system? What make/brand are the GPUs? The GPU boost and fan control may differ from brand to brand and affect how hard each GPU is being pushed/throttled. – Romen Sep 10 at 16:33
  • Both the GPU were tested in same system. – israfil ansari Sep 11 at 4:39
  • 1
    I have solved the problem by "compiling darknet library with GPU=1 CUDNN=1 CUDNN_HALF=0 OPENCV=1" – israfil ansari Sep 11 at 6:03
  • You can add that as an answer to your own question, or alternatively remove it since this is off-topic for Hardware Recommendations. – Romen Sep 12 at 16:56

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.