OpenCV Cuda with GPU which does not support CUDA

1

Hi,

I have a GT1030 GPU on my desktop setup. I checked official NVIDIA page and it seems like GT1030 does not support Cuda. Without GPU Cuda support, can i use OpenCV Cuda modules?

My other question is can i use opencv cuda modules on my macbook air(intel chip, big sur)? During installation of OpenCV, there is a CMake flags for CUDA on Linux tutorial. But i did not see on MacOS. Does same flags apply to macos installation too?

2

I see no reason to think that.

I see claims that it supports compute capability 6.1, which is representative of the GeForce 10 series.

it has a modest amount of GPU RAM and less than 1 Tflop/s of FP32 performance but it should do the job.

no, because CUDA equals NVIDIA and that hardware doesn’t contain an NVIDIA GPU.

instead you can use OpenCL. OpenCV’s DNN module has an OpenCL backend. many non-DNN functions in OpenCV have an OpenCL acceleration path as well.

3

Thank you for quick reply.

My final product is Jetson Nano so i dont want to use OpenCL during development.(i am not sure if this decision is right :slight_smile: )

I will use my desktop then.

I checked this link to see if my GPU supports Cuda. But i did not see. That was why i said my GPU does not support Cuda.

4

According to the Nvidia spec sheet (cached) the GT1030 has 384 CUDA cores. So it supports CUDA parallel processing.

You didn’t tell what do you want to use GPU processing for. Most OpenCV functions have CUDA and OpenCL backend too, so you can use both. If you want to develop your own image processing functions on GPU, it’s another problem, and you should probably look for help in another forum.

Anyway, CUDA and OpenCL are doing basically the same thing (running massively parallel operations on the GPU), the difference is that CUDA works only on Nvidia GPUs, OpenCL runs on most GPUs and also on CPUs. It also seems that CUDA development is easier and better optimized (this is only a personal impression).

For GPU development I suggest to use your desktop computer rather than a Macbook Air or a NVidia Nano (which is good for CUDA processing, but slow in compiling).

1 Like