{
  "id": 100935,
  "title": "[naive request] more cpu cores for gpu instances?",
  "url": "/competitions/recursion-cellular-image-classification/discussion/100935",
  "author_name": "",
  "post_date": "2019-07-22T08:14:07.486837Z",
  "votes": 9,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I believe many competitors feel a cpu bottleneck here and it looks like gpu idle time is around 90-95%. So quite naive request to kaggle team: any chance to update gpu instances with few more cores/threads? </p>",
  "messages": [
    {
      "id": "581653",
      "postDate": "07/22/2019 08:14:07",
      "content": "<p>I believe many competitors feel a cpu bottleneck here and it looks like gpu idle time is around 90-95%. So quite naive request to kaggle team: any chance to update gpu instances with few more cores/threads? </p>",
      "rawMarkdown": "I believe many competitors feel a cpu bottleneck here and it looks like gpu idle time is around 90-95%. So quite naive request to kaggle team: any chance to update gpu instances with few more cores/threads?",
      "votes": null
    },
    {
      "id": "582248",
      "postDate": "07/23/2019 00:23:16",
      "content": "<p>There are hardware limitations, You have a certain number of cores in a machine with a certain number of GPU's. If you give one GPU more cores then it takes away from other GPU's. </p>",
      "rawMarkdown": "There are hardware limitations, You have a certain number of cores in a machine with a certain number of GPU's. If you give one GPU more cores then it takes away from other GPU's.",
      "votes": null
    },
    {
      "id": "582810",
      "postDate": "07/23/2019 16:12:38",
      "content": "<p>I think it’s just matter of kaggle resources and for sure one day they will upgrade gpu kernels, but that would be great for this competition specially.</p>",
      "rawMarkdown": "I think it’s just matter of kaggle resources and for sure one day they will upgrade gpu kernels, but that would be great for this competition specially.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 582248,
      "author_name": "cherring",
      "author_url": "",
      "post_date": "07/23/2019 00:23:16",
      "content": "<p>There are hardware limitations, You have a certain number of cores in a machine with a certain number of GPU's. If you give one GPU more cores then it takes away from other GPU's. </p>",
      "votes": null,
      "replies": [
        {
          "id": 582810,
          "author_name": "yaroshevskiy",
          "author_url": "",
          "post_date": "07/23/2019 16:12:38",
          "content": "<p>I think it’s just matter of kaggle resources and for sure one day they will upgrade gpu kernels, but that would be great for this competition specially.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "581653": "I believe many competitors feel a cpu bottleneck here and it looks like gpu idle time is around 90-95%. So quite naive request to kaggle team: any chance to update gpu instances with few more cores/threads?",
    "582248": "There are hardware limitations, You have a certain number of cores in a machine with a certain number of GPU's. If you give one GPU more cores then it takes away from other GPU's.",
    "582810": "I think it’s just matter of kaggle resources and for sure one day they will upgrade gpu kernels, but that would be great for this competition specially."
  },
  "source": "meta"
}