{
  "id": 346625,
  "title": "I should buy a 5995wx and 4090 GPU after this competition",
  "url": "/competitions/amex-default-prediction/discussion/346625",
  "author_name": "bjjiang",
  "post_date": "2022-08-20T15:23:01.179000",
  "votes": 5,
  "comment_count": 8,
  "views": 0,
  "content": "<p>A lot of idea consist in my head but I can't try them. The problem lies in training speed. It was too late for me to join in this competition, just 3 weeks ago</p>",
  "messages": [
    {
      "id": 1907214,
      "postDate": "2022-08-20T15:23:01.180Z",
      "content": "<p>A lot of idea consist in my head but I can't try them. The problem lies in training speed. It was too late for me to join in this competition, just 3 weeks ago</p>",
      "rawMarkdown": "A lot of idea consist in my head but I can't try them. The problem lies in training speed. It was too late for me to join in this competition, just 3 weeks ago",
      "votes": 5
    },
    {
      "id": 1907325,
      "postDate": "2022-08-20T17:08:38.313Z",
      "content": "<p>I guess it's always good to have a high config machine if you intend to continue competing in Kaggle. The competitions these days have become quite computationally intensive. For example, loosely speaking, in NLP and Computer Vision comps, you will usually be better off training a large model than a base version, and you will need extra ram to host a large model. </p>",
      "rawMarkdown": "I guess it's always good to have a high config machine if you intend to continue competing in Kaggle. The competitions these days have become quite computationally intensive. For example, loosely speaking, in NLP and Computer Vision comps, you will usually be better off training a large model than a base version, and you will need extra ram to host a large model. ",
      "votes": 4
    },
    {
      "id": 1907230,
      "postDate": "2022-08-20T15:40:55.007Z",
      "content": "<p>For me, 2x 3060 (12GB) is enough to do anything </p>",
      "rawMarkdown": "For me, 2x 3060 (12GB) is enough to do anything ",
      "votes": 1,
      "replies": [
        {
          "id": 1907266,
          "postDate": "2022-08-20T16:10:52.523Z",
          "content": "<p>my 1070 can‘t work in lightgbm， I don't know why</p>",
          "rawMarkdown": "my 1070 can‘t work in lightgbm， I don't know why",
          "votes": 1
        },
        {
          "id": 1907271,
          "postDate": "2022-08-20T16:16:23.463Z",
          "content": "<p>I think because you had the problem with memory, 12GB in the GPU maybe it's ok, but the double (12x2) is more efficient for everything you need to do </p>",
          "rawMarkdown": "I think because you had the problem with memory, 12GB in the GPU maybe it's ok, but the double (12x2) is more efficient for everything you need to do "
        },
        {
          "id": 1907497,
          "postDate": "2022-08-20T20:55:32.133Z",
          "content": "<blockquote>\n  <p>my 1070 can‘t work in lightgbm， I don't know why</p>\n</blockquote>\n<p><a href=\"https://www.kaggle.com/bjjiang\" target=\"_blank\">@bjjiang</a> a GPU often doesn't do much to speed up lightgbm compared to running lightgbm on a CPU. Xgboost on the other hand is much faster on GPU.</p>",
          "rawMarkdown": "> my 1070 can‘t work in lightgbm， I don't know why\n\n@bjjiang a GPU often doesn't do much to speed up lightgbm compared to running lightgbm on a CPU. Xgboost on the other hand is much faster on GPU.",
          "votes": 2
        }
      ]
    },
    {
      "id": 1907612,
      "postDate": "2022-08-20T23:55:23.740Z",
      "content": "<p>Hello, <a href=\"https://www.kaggle.com/bjjiang\" target=\"_blank\">@bjjiang</a>. That's a significant investment; have you tried to use any cloud environment like GCP, Colab Pro+, Azure or AWS, and many more?</p>",
      "rawMarkdown": "Hello, @bjjiang. That's a significant investment; have you tried to use any cloud environment like GCP, Colab Pro+, Azure or AWS, and many more?\n",
      "replies": [
        {
          "id": 1907666,
          "postDate": "2022-08-21T00:48:31.410Z",
          "rawMarkdown": "",
          "votes": 1,
          "isDeleted": true
        }
      ]
    },
    {
      "id": 1907221,
      "postDate": "2022-08-20T15:31:39.257Z",
      "rawMarkdown": "",
      "votes": -1,
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 1907325,
      "author_name": "Tonghui Li",
      "author_url": "",
      "post_date": "2022-08-20T17:08:38.313000",
      "content": "<p>I guess it's always good to have a high config machine if you intend to continue competing in Kaggle. The competitions these days have become quite computationally intensive. For example, loosely speaking, in NLP and Computer Vision comps, you will usually be better off training a large model than a base version, and you will need extra ram to host a large model. </p>",
      "votes": 4,
      "replies": []
    },
    {
      "id": 1907230,
      "author_name": "EL Younes",
      "author_url": "",
      "post_date": "2022-08-20T15:40:55.007000",
      "content": "<p>For me, 2x 3060 (12GB) is enough to do anything </p>",
      "votes": 1,
      "replies": [
        {
          "id": 1907266,
          "author_name": "bjjiang",
          "author_url": "",
          "post_date": "2022-08-20T16:10:52.523000",
          "content": "<p>my 1070 can‘t work in lightgbm， I don't know why</p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 1907271,
          "author_name": "EL Younes",
          "author_url": "",
          "post_date": "2022-08-20T16:16:23.463000",
          "content": "<p>I think because you had the problem with memory, 12GB in the GPU maybe it's ok, but the double (12x2) is more efficient for everything you need to do </p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 1907497,
          "author_name": "MichaelP",
          "author_url": "",
          "post_date": "2022-08-20T20:55:32.133000",
          "content": "<blockquote>\n  <p>my 1070 can‘t work in lightgbm， I don't know why</p>\n</blockquote>\n<p><a href=\"https://www.kaggle.com/bjjiang\" target=\"_blank\">@bjjiang</a> a GPU often doesn't do much to speed up lightgbm compared to running lightgbm on a CPU. Xgboost on the other hand is much faster on GPU.</p>",
          "votes": 2,
          "replies": []
        }
      ]
    },
    {
      "id": 1907612,
      "author_name": "C4rl05/V",
      "author_url": "",
      "post_date": "2022-08-20T23:55:23.740000",
      "content": "<p>Hello, <a href=\"https://www.kaggle.com/bjjiang\" target=\"_blank\">@bjjiang</a>. That's a significant investment; have you tried to use any cloud environment like GCP, Colab Pro+, Azure or AWS, and many more?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 1907666,
          "author_name": "",
          "author_url": "",
          "post_date": "2022-08-21T00:48:31.410000",
          "content": "",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 1907221,
      "author_name": "",
      "author_url": "",
      "post_date": "2022-08-20T15:31:39.257000",
      "content": "",
      "votes": -1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1907214": "A lot of idea consist in my head but I can't try them. The problem lies in training speed. It was too late for me to join in this competition, just 3 weeks ago",
    "1907325": "I guess it's always good to have a high config machine if you intend to continue competing in Kaggle. The competitions these days have become quite computationally intensive. For example, loosely speaking, in NLP and Computer Vision comps, you will usually be better off training a large model than a base version, and you will need extra ram to host a large model. ",
    "1907230": "For me, 2x 3060 (12GB) is enough to do anything ",
    "1907612": "Hello, @bjjiang. That's a significant investment; have you tried to use any cloud environment like GCP, Colab Pro+, Azure or AWS, and many more?\n",
    "1907221": ""
  }
}