Running CUDA enabled computer vision code on windows

I am working on a computer vision project. I am planning on using CUDA. I am running a Windows 10 laptop with Nvidia GTX 960m and Visual Studio 2015 Community Edition. My final objective is to get images from the cameras and process them on my laptop using the GPU.
Problem is Windows provides two driver models under which Nvidia driver may operate http://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html#driver-model, the WDDM Driver model and TCC mode (not supported by Geforce GTX 960m). This will probably cause Timeout problems when running the kernel modules on the GPU.
Notes:

  • http://stackoverflow.com/questions/15344483/cuda-on-windows-and-linux: This post suggests that I can debug CUDA code on Windows but nothing about running CUDA programs or the Timeout issues is mentioned.
  • https://devtalk.nvidia.com/default/topic/490168/windows-or-linux-for-cuda/: This post points out the timeout issues on windows and needing to run Linux without GUI while debugging. But its pretty old - 2011.
  • http://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html#driver-model: It is also mentioned, that WDDM Driver model is used for display devices. But on my system, the Intel integrated graphics is used for display purposes under normal conditions (when not doing anything GPU intensive). See this image: https://drive.google.com/open?id=0B9dJ1h0j_vMDdzRBMXFCeVpVMkk
  • So, Would It be possible do the image processing on Windows on the GPU? I want to make sure that CUDA will fulfill my needs before installing it on my system.

    Hi, harshatech2012

    Firstly if you want to do cuda debug on VS, you should install Nsight VSE.
    https://developer.nvidia.com/nvidia-nsight-visual-studio-edition

    Then before do cuda-debugging, you should increase the TDR by set Nsight Monitor options->General->WDDM TDR Delay to a larger number, like 300s

    Then suppose you can debug your app.

    Ok. Thank you.

    I have one more thing to ask, but this might be out of the context of the current discussion. I would like to know how to update CUDA toolkit?
    Now the current relevant version for me is labeled “cuda_8.0.44_win10.exe”. What should I do if a newer version, say “cuda_8.0.47_win10.exe” is released.? I searched a lot, but couldn’t find anything regarding updating the toolkit.

    Hi, harshatech2012

    Just install the new exe file can update the toolkit.
    After install, you can “nvprof --version” in cmd to check if the version has updated

    Nvidia Nsight Studio only supports certain graphics cards (Link:https://developer.nvidia.com/nsight-visual-studio-edition-supported-gpus-full-list). And GTX 960M is not included in the complete list. So, would it work on my system as I only have a GTX 960M.?

    Basically, it should be supported.