{
  "id": 130536,
  "title": "Low GPU utilization in kernel",
  "url": "/competitions/deepfake-detection-challenge/discussion/130536",
  "author_name": "",
  "post_date": "2020-02-14T18:29:07.039395500Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi, \nI enabled GPU accelerator on my kernel and while training, GPU utilization is 0% while CPU utilization is exceeding 100% (as seen in the picture below).\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4420071%2F61b0da92410abb90983a3f62a4aabe52%2FCapture.PNG?generation=1581622773674781&amp;alt=media\" alt=\"\">\nThis is the part of code for which I've taken the above snap. It is taking around 67 seconds (even with GPU (utilization = 0%)) for 1 loop (1 loop performs capturing 100 frames and writing them) and there are 400 loops in total (As 400 videos). Does OpenCV not work with GPU? Pls help!\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4420071%2F34c5fa1c0f8399642db9a5f814098aa7%2FCapture.PNG?generation=1581649203115732&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": "746204",
      "postDate": "02/14/2020 18:29:07",
      "content": "<p>Hi, \nI enabled GPU accelerator on my kernel and while training, GPU utilization is 0% while CPU utilization is exceeding 100% (as seen in the picture below).\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4420071%2F61b0da92410abb90983a3f62a4aabe52%2FCapture.PNG?generation=1581622773674781&amp;alt=media\" alt=\"\">\nThis is the part of code for which I've taken the above snap. It is taking around 67 seconds (even with GPU (utilization = 0%)) for 1 loop (1 loop performs capturing 100 frames and writing them) and there are 400 loops in total (As 400 videos). Does OpenCV not work with GPU? Pls help!\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4420071%2F34c5fa1c0f8399642db9a5f814098aa7%2FCapture.PNG?generation=1581649203115732&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "Hi, \nI enabled GPU accelerator on my kernel and while training, GPU utilization is 0% while CPU utilization is exceeding 100% (as seen in the picture below).\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4420071%2F61b0da92410abb90983a3f62a4aabe52%2FCapture.PNG?generation=1581622773674781&amp;alt=media)\nThis is the part of code for which I've taken the above snap. It is taking around 67 seconds (even with GPU (utilization = 0%)) for 1 loop (1 loop performs capturing 100 frames and writing them) and there are 400 loops in total (As 400 videos). Does OpenCV not work with GPU? Pls help!\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4420071%2F34c5fa1c0f8399642db9a5f814098aa7%2FCapture.PNG?generation=1581649203115732&amp;alt=media)",
      "votes": null
    },
    {
      "id": "746248",
      "postDate": "02/14/2020 19:30:30",
      "content": "<p>There are OpenCV modules exploiting the GPU but not that one as far as I know. You might want to look for ffmpeg, in particular the command enabling the GPU as well. I haven't tried doing that in the past so I don't know how much that can help.</p>",
      "rawMarkdown": "There are OpenCV modules exploiting the GPU but not that one as far as I know. You might want to look for ffmpeg, in particular the command enabling the GPU as well. I haven't tried doing that in the past so I don't know how much that can help.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 746248,
      "author_name": "morrizzz",
      "author_url": "",
      "post_date": "02/14/2020 19:30:30",
      "content": "<p>There are OpenCV modules exploiting the GPU but not that one as far as I know. You might want to look for ffmpeg, in particular the command enabling the GPU as well. I haven't tried doing that in the past so I don't know how much that can help.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "746204": "Hi, \nI enabled GPU accelerator on my kernel and while training, GPU utilization is 0% while CPU utilization is exceeding 100% (as seen in the picture below).\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4420071%2F61b0da92410abb90983a3f62a4aabe52%2FCapture.PNG?generation=1581622773674781&amp;alt=media)\nThis is the part of code for which I've taken the above snap. It is taking around 67 seconds (even with GPU (utilization = 0%)) for 1 loop (1 loop performs capturing 100 frames and writing them) and there are 400 loops in total (As 400 videos). Does OpenCV not work with GPU? Pls help!\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F4420071%2F34c5fa1c0f8399642db9a5f814098aa7%2FCapture.PNG?generation=1581649203115732&amp;alt=media)",
    "746248": "There are OpenCV modules exploiting the GPU but not that one as far as I know. You might want to look for ffmpeg, in particular the command enabling the GPU as well. I haven't tried doing that in the past so I don't know how much that can help."
  },
  "source": "meta"
}