{
  "id": 104255,
  "title": "GPU out of memory",
  "url": "/competitions/aptos2019-blindness-detection/discussion/104255",
  "author_name": "[ods.ai] Kyrylo",
  "post_date": "2019-08-15T14:46:14.102000",
  "votes": 0,
  "comment_count": 5,
  "views": 0,
  "content": "<p>I tried to train my model locally because Kaggle doesn't offer a good run time. I ran into the following error.(I am using fastai)\n<code>CUDA out of memory. Tried to allocate 1.13 GiB (GPU 0; 8.00 GiB total capacity; 564.45 MiB already allocated; 5.42 GiB free; 55.55 MiB cached)</code>\nThis error is very confusing because i have 5.42 GIB free and 1.13 should not be a problem. I tried to decrease my batch size but it still doesn't  seem to work. </p>",
  "messages": [
    {
      "id": 602066,
      "postDate": "2019-08-18T14:18:45.967Z",
      "content": "<p>Does CUDA or your operating system set a limit to how much GPU memory you can utilize when running your pipeline? It could be that a lot of the GPU memory is assigned to a different task (for example an extra screen or visual program) and that the operating system doesn't allow more to be allocated.</p>\n\n<p><a href=\"https://itstillworks.com/do-allocate-power-graphics-card-6018217.html\">https://itstillworks.com/do-allocate-power-graphics-card-6018217.html</a></p>\n\n<p>Hope this helps!</p>",
      "rawMarkdown": "Does CUDA or your operating system set a limit to how much GPU memory you can utilize when running your pipeline? It could be that a lot of the GPU memory is assigned to a different task (for example an extra screen or visual program) and that the operating system doesn't allow more to be allocated.\n\nhttps://itstillworks.com/do-allocate-power-graphics-card-6018217.html\n\nHope this helps!",
      "votes": 1
    },
    {
      "id": 600023,
      "postDate": "2019-08-15T14:46:14.103Z",
      "content": "<p>I tried to train my model locally because Kaggle doesn't offer a good run time. I ran into the following error.(I am using fastai)\n<code>CUDA out of memory. Tried to allocate 1.13 GiB (GPU 0; 8.00 GiB total capacity; 564.45 MiB already allocated; 5.42 GiB free; 55.55 MiB cached)</code>\nThis error is very confusing because i have 5.42 GIB free and 1.13 should not be a problem. I tried to decrease my batch size but it still doesn't  seem to work. </p>",
      "rawMarkdown": "I tried to train my model locally because Kaggle doesn't offer a good run time. I ran into the following error.(I am using fastai)\n`CUDA out of memory. Tried to allocate 1.13 GiB (GPU 0; 8.00 GiB total capacity; 564.45 MiB already allocated; 5.42 GiB free; 55.55 MiB cached)`\nThis error is very confusing because i have 5.42 GIB free and 1.13 should not be a problem. I tried to decrease my batch size but it still doesn't  seem to work. "
    },
    {
      "id": 606327,
      "postDate": "2019-08-23T12:50:43.923Z",
      "rawMarkdown": "",
      "isDeleted": true
    },
    {
      "id": 602243,
      "postDate": "2019-08-18T20:10:49.197Z",
      "rawMarkdown": "",
      "isDeleted": true
    },
    {
      "id": 601312,
      "postDate": "2019-08-17T12:01:38.867Z",
      "rawMarkdown": "",
      "votes": 1,
      "isDeleted": true
    },
    {
      "id": 600167,
      "postDate": "2019-08-15T18:36:44.543Z",
      "rawMarkdown": "",
      "votes": 2,
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 602066,
      "author_name": "Carlo",
      "author_url": "",
      "post_date": "2019-08-18T14:18:45.967000",
      "content": "<p>Does CUDA or your operating system set a limit to how much GPU memory you can utilize when running your pipeline? It could be that a lot of the GPU memory is assigned to a different task (for example an extra screen or visual program) and that the operating system doesn't allow more to be allocated.</p>\n\n<p><a href=\"https://itstillworks.com/do-allocate-power-graphics-card-6018217.html\">https://itstillworks.com/do-allocate-power-graphics-card-6018217.html</a></p>\n\n<p>Hope this helps!</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 606327,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-08-23T12:50:43.923000",
      "content": "",
      "votes": 0,
      "replies": []
    },
    {
      "id": 602243,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-08-18T20:10:49.197000",
      "content": "",
      "votes": 0,
      "replies": []
    },
    {
      "id": 601312,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-08-17T12:01:38.867000",
      "content": "",
      "votes": 1,
      "replies": []
    },
    {
      "id": 600167,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-08-15T18:36:44.543000",
      "content": "",
      "votes": 2,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "602066": "Does CUDA or your operating system set a limit to how much GPU memory you can utilize when running your pipeline? It could be that a lot of the GPU memory is assigned to a different task (for example an extra screen or visual program) and that the operating system doesn't allow more to be allocated.\n\nhttps://itstillworks.com/do-allocate-power-graphics-card-6018217.html\n\nHope this helps!",
    "600023": "I tried to train my model locally because Kaggle doesn't offer a good run time. I ran into the following error.(I am using fastai)\n`CUDA out of memory. Tried to allocate 1.13 GiB (GPU 0; 8.00 GiB total capacity; 564.45 MiB already allocated; 5.42 GiB free; 55.55 MiB cached)`\nThis error is very confusing because i have 5.42 GIB free and 1.13 should not be a problem. I tried to decrease my batch size but it still doesn't  seem to work. ",
    "606327": "",
    "602243": "",
    "601312": "",
    "600167": ""
  }
}