{
  "id": 180947,
  "title": "CUDA : out of memory",
  "url": "/competitions/landmark-recognition-2020/discussion/180947",
  "author_name": "",
  "post_date": "2020-09-07T02:53:52.296130200Z",
  "votes": 3,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Help needed in resolving this error of CUDA, out of memory</p>",
  "messages": [
    {
      "id": "1001029",
      "postDate": "09/07/2020 02:53:52",
      "content": "<p>Help needed in resolving this error of CUDA, out of memory</p>",
      "rawMarkdown": "Help needed in resolving this error of CUDA, out of memory",
      "votes": null
    },
    {
      "id": "1001079",
      "postDate": "09/07/2020 04:08:11",
      "content": "<p>Reduce Batch Size, CUDA is getting out of memory due to images, and other things which are taking a lot of memory. </p>\n<p>Try to reduce the memory getting used, batch size is just one thing.</p>",
      "rawMarkdown": "Reduce Batch Size, CUDA is getting out of memory due to images, and other things which are taking a lot of memory. \n\nTry to reduce the memory getting used, batch size is just one thing.",
      "votes": null
    },
    {
      "id": "1001139",
      "postDate": "09/07/2020 05:04:37",
      "content": "<p>I tried <a href=\"https://www.kaggle.com/sarques\" target=\"_blank\">@sarques</a>. But no luck</p>",
      "rawMarkdown": "I tried @sarques. But no luck",
      "votes": null
    },
    {
      "id": "1001146",
      "postDate": "09/07/2020 05:13:07",
      "content": "<p>Try with a different<strong>batch size like reduce the batch size to 12,8</strong> and so on, but training on lower batch size will take more time, and you can do one more thing like reducing the memory taken by the feature like <strong>convert all the float 64 to float 32 and int 32 to int 16 and so on.</strong><br>\nBy using this strategy you can reduce the memory utilized by the features.<br>\nHope this helps you little bit</p>",
      "rawMarkdown": "Try with a different**batch size like reduce the batch size to 12,8** and so on, but training on lower batch size will take more time, and you can do one more thing like reducing the memory taken by the feature like **convert all the float 64 to float 32 and int 32 to int 16 and so on.**\nBy using this strategy you can reduce the memory utilized by the features.\nHope this helps you little bit",
      "votes": null
    },
    {
      "id": "1001180",
      "postDate": "09/07/2020 05:55:01",
      "content": "<p>Thank you <a href=\"https://www.kaggle.com/vpkprasanna\" target=\"_blank\">@vpkprasanna</a> </p>",
      "rawMarkdown": "Thank you @vpkprasanna",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1001079,
      "author_name": "sarques",
      "author_url": "",
      "post_date": "09/07/2020 04:08:11",
      "content": "<p>Reduce Batch Size, CUDA is getting out of memory due to images, and other things which are taking a lot of memory. </p>\n<p>Try to reduce the memory getting used, batch size is just one thing.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1001139,
      "author_name": "sanjaydsb",
      "author_url": "",
      "post_date": "09/07/2020 05:04:37",
      "content": "<p>I tried <a href=\"https://www.kaggle.com/sarques\" target=\"_blank\">@sarques</a>. But no luck</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1001146,
      "author_name": "vpkprasanna",
      "author_url": "",
      "post_date": "09/07/2020 05:13:07",
      "content": "<p>Try with a different<strong>batch size like reduce the batch size to 12,8</strong> and so on, but training on lower batch size will take more time, and you can do one more thing like reducing the memory taken by the feature like <strong>convert all the float 64 to float 32 and int 32 to int 16 and so on.</strong><br>\nBy using this strategy you can reduce the memory utilized by the features.<br>\nHope this helps you little bit</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1001180,
      "author_name": "sanjaydsb",
      "author_url": "",
      "post_date": "09/07/2020 05:55:01",
      "content": "<p>Thank you <a href=\"https://www.kaggle.com/vpkprasanna\" target=\"_blank\">@vpkprasanna</a> </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1001029": "Help needed in resolving this error of CUDA, out of memory",
    "1001079": "Reduce Batch Size, CUDA is getting out of memory due to images, and other things which are taking a lot of memory. \n\nTry to reduce the memory getting used, batch size is just one thing.",
    "1001139": "I tried @sarques. But no luck",
    "1001146": "Try with a different**batch size like reduce the batch size to 12,8** and so on, but training on lower batch size will take more time, and you can do one more thing like reducing the memory taken by the feature like **convert all the float 64 to float 32 and int 32 to int 16 and so on.**\nBy using this strategy you can reduce the memory utilized by the features.\nHope this helps you little bit",
    "1001180": "Thank you @vpkprasanna"
  },
  "source": "meta"
}