{
  "id": 127912,
  "title": "How to get OpenCV GPU locally?",
  "url": "/competitions/deepfake-detection-challenge/discussion/127912",
  "author_name": "",
  "post_date": "2020-01-27T16:09:23.361244200Z",
  "votes": null,
  "comment_count": 4,
  "views": 0,
  "content": "<p>How does Kaggle run OpenCV faster on GPU notebook? Can I build openCV a certain way on my local machine so it runs faster like Kaggle? I see a few stackoverflow questions which say you can build a wrapper, but Kaggle does it with no changes?</p>",
  "messages": [
    {
      "id": "730551",
      "postDate": "01/27/2020 16:09:23",
      "content": "<p>How does Kaggle run OpenCV faster on GPU notebook? Can I build openCV a certain way on my local machine so it runs faster like Kaggle? I see a few stackoverflow questions which say you can build a wrapper, but Kaggle does it with no changes?</p>",
      "rawMarkdown": "How does Kaggle run OpenCV faster on GPU notebook? Can I build openCV a certain way on my local machine so it runs faster like Kaggle? I see a few stackoverflow questions which say you can build a wrapper, but Kaggle does it with no changes?",
      "votes": null
    },
    {
      "id": "730649",
      "postDate": "01/27/2020 18:22:43",
      "content": "<p>You compile opencv with gpu support <a href=\"http://www.pyimagesearch.com/2016/07/11/compiling-opencv-with-cuda-support/\">link</a>. Kaggle does it with no change is probably because of it have already been compiled(docker). You can check by <code>print(cv2.getBuildInformation())</code></p>",
      "rawMarkdown": "You compile opencv with gpu support [link](http://www.pyimagesearch.com/2016/07/11/compiling-opencv-with-cuda-support/). Kaggle does it with no change is probably because of it have already been compiled(docker). You can check by `print(cv2.getBuildInformation())`",
      "votes": null
    },
    {
      "id": "730659",
      "postDate": "01/27/2020 18:45:36",
      "content": "<p>Thanks for link I will try!</p>",
      "rawMarkdown": "Thanks for link I will try!",
      "votes": null
    },
    {
      "id": "731559",
      "postDate": "01/28/2020 19:33:36",
      "content": "<p>I install and see  </p>\n\n<pre><code>NVIDIA CUDA:                 YES (ver 10.1, CUFFT CUBLAS FAST_MATH)\nNVIDIA GPU arch:             53 60 61 70 75\nNVIDIA PTX archs:            75\n\ncuDNN:                       YES (ver 7.6.4)\n</code></pre>\n\n<p>But it doesn't seem like there was any gain. I think kaggle must have something different as well.</p>",
      "rawMarkdown": "I install and see  \n    \n    NVIDIA CUDA:                 YES (ver 10.1, CUFFT CUBLAS FAST_MATH)\n    NVIDIA GPU arch:             53 60 61 70 75\n    NVIDIA PTX archs:            75\n\n    cuDNN:                       YES (ver 7.6.4)\n\nBut it doesn't seem like there was any gain. I think kaggle must have something different as well.",
      "votes": null
    },
    {
      "id": "731580",
      "postDate": "01/28/2020 19:52:11",
      "content": "<p>I go here: <a href=\"https://github.com/Kaggle/docker-python/blob/master/gpu.Dockerfile\">https://github.com/Kaggle/docker-python/blob/master/gpu.Dockerfile</a> and see they have </p>\n\n<pre><code>RUN pip install pycuda &amp;&amp; \\\npip install cupy-cuda100 &amp;&amp; \\\npip install pynvrtc &amp;&amp; \\\n/tmp/clean-layer.sh\n</code></pre>\n\n<p>so I will try these and see if it changes anything</p>",
      "rawMarkdown": "I go here: https://github.com/Kaggle/docker-python/blob/master/gpu.Dockerfile and see they have \n    \n    RUN pip install pycuda &amp;&amp; \\\n    pip install cupy-cuda100 &amp;&amp; \\\n    pip install pynvrtc &amp;&amp; \\\n    /tmp/clean-layer.sh\n\nso I will try these and see if it changes anything",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 730649,
      "author_name": "unkownhihi",
      "author_url": "",
      "post_date": "01/27/2020 18:22:43",
      "content": "<p>You compile opencv with gpu support <a href=\"http://www.pyimagesearch.com/2016/07/11/compiling-opencv-with-cuda-support/\">link</a>. Kaggle does it with no change is probably because of it have already been compiled(docker). You can check by <code>print(cv2.getBuildInformation())</code></p>",
      "votes": null,
      "replies": [
        {
          "id": 730659,
          "author_name": "sethkitchen",
          "author_url": "",
          "post_date": "01/27/2020 18:45:36",
          "content": "<p>Thanks for link I will try!</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 731559,
          "author_name": "sethkitchen",
          "author_url": "",
          "post_date": "01/28/2020 19:33:36",
          "content": "<p>I install and see  </p>\n\n<pre><code>NVIDIA CUDA:                 YES (ver 10.1, CUFFT CUBLAS FAST_MATH)\nNVIDIA GPU arch:             53 60 61 70 75\nNVIDIA PTX archs:            75\n\ncuDNN:                       YES (ver 7.6.4)\n</code></pre>\n\n<p>But it doesn't seem like there was any gain. I think kaggle must have something different as well.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 731580,
      "author_name": "sethkitchen",
      "author_url": "",
      "post_date": "01/28/2020 19:52:11",
      "content": "<p>I go here: <a href=\"https://github.com/Kaggle/docker-python/blob/master/gpu.Dockerfile\">https://github.com/Kaggle/docker-python/blob/master/gpu.Dockerfile</a> and see they have </p>\n\n<pre><code>RUN pip install pycuda &amp;&amp; \\\npip install cupy-cuda100 &amp;&amp; \\\npip install pynvrtc &amp;&amp; \\\n/tmp/clean-layer.sh\n</code></pre>\n\n<p>so I will try these and see if it changes anything</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "730551": "How does Kaggle run OpenCV faster on GPU notebook? Can I build openCV a certain way on my local machine so it runs faster like Kaggle? I see a few stackoverflow questions which say you can build a wrapper, but Kaggle does it with no changes?",
    "730649": "You compile opencv with gpu support [link](http://www.pyimagesearch.com/2016/07/11/compiling-opencv-with-cuda-support/). Kaggle does it with no change is probably because of it have already been compiled(docker). You can check by `print(cv2.getBuildInformation())`",
    "730659": "Thanks for link I will try!",
    "731559": "I install and see  \n    \n    NVIDIA CUDA:                 YES (ver 10.1, CUFFT CUBLAS FAST_MATH)\n    NVIDIA GPU arch:             53 60 61 70 75\n    NVIDIA PTX archs:            75\n\n    cuDNN:                       YES (ver 7.6.4)\n\nBut it doesn't seem like there was any gain. I think kaggle must have something different as well.",
    "731580": "I go here: https://github.com/Kaggle/docker-python/blob/master/gpu.Dockerfile and see they have \n    \n    RUN pip install pycuda &amp;&amp; \\\n    pip install cupy-cuda100 &amp;&amp; \\\n    pip install pynvrtc &amp;&amp; \\\n    /tmp/clean-layer.sh\n\nso I will try these and see if it changes anything"
  },
  "source": "meta"
}