{
  "id": 666350,
  "title": "Could Kaggle provide one submission per day even running out gpu quota?",
  "url": "/competitions/vesuvius-challenge-surface-detection/discussion/666350",
  "author_name": "",
  "post_date": "2026-01-06T15:19:20.174566500Z",
  "votes": -5,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Running out of GPU due to training,  submission is impossible anymore.\nIs it possible to provide one submission  per day?</p>",
  "messages": [
    {
      "id": "3387219",
      "postDate": "01/06/2026 15:19:20",
      "content": "<p>Running out of GPU due to training,  submission is impossible anymore.\nIs it possible to provide one submission  per day?</p>",
      "rawMarkdown": "Running out of GPU due to training,  submission is impossible anymore.\nIs it possible to provide one submission  per day?",
      "votes": null
    },
    {
      "id": "3387440",
      "postDate": "01/07/2026 02:45:54",
      "content": "<p>I always run out of my gpu quota in 3 days, But I always 2-4 hours as backup, I think you should also start doing this from next week.</p>",
      "rawMarkdown": "I always run out of my gpu quota in 3 days, But I always 2-4 hours as backup, I think you should also start doing this from next week.",
      "votes": null
    },
    {
      "id": "3387446",
      "postDate": "01/07/2026 03:04:07",
      "content": "<p>Ideally, I always have 30 mins of buffer GPU quota for a week which I plan to keep unused. If I reach close to the gpu weekly limit, I do the public LB sub first and the validation experiment later.</p>\n<p>But I try to prioritize training on TPU as much as possible but even that proves to be difficult sometimes due to the extensive queue for TPU's (sometimes 2-3 hours lol)</p>",
      "rawMarkdown": "Ideally, I always have 30 mins of buffer GPU quota for a week which I plan to keep unused. If I reach close to the gpu weekly limit, I do the public LB sub first and the validation experiment later.\n\nBut I try to prioritize training on TPU as much as possible but even that proves to be difficult sometimes due to the extensive queue for TPU's (sometimes 2-3 hours lol)",
      "votes": null
    },
    {
      "id": "3387746",
      "postDate": "01/07/2026 15:20:16",
      "content": "<p>Thanks for all reply.   IT takes relatively long for training.  If I reserve submission gpu, then I can't finish my training. Anyway, resources limited.</p>",
      "rawMarkdown": "Thanks for all reply.   IT takes relatively long for training.  If I reserve submission gpu, then I can't finish my training. Anyway, resources limited.",
      "votes": null
    },
    {
      "id": "3388034",
      "postDate": "01/08/2026 02:24:54",
      "content": "<p>One option is to buy a Google Colab subscription for $10 USD / month. If you link your Kaggle account to your Colab account you're granted another 15 hours weekly. You also end up with the extra colab hours that you get per month with the plan. Given that's around 60 hours 2xT4 extra for the Pro plan per month for $10 it's a good deal if you're actively utilizing the hours.</p>\n<p><a href=\"https://www.kaggle.com/blog/level-up-your-compute-more-gpu-hours-on-kaggle-wit\" target=\"_blank\">https://www.kaggle.com/blog/level-up-your-compute-more-gpu-hours-on-kaggle-wit</a></p>\n<blockquote>\n  <p>Colab Pro subscribers: Get an additional 15 GPU hours per week.\n     Colab Pro+ subscribers: Get an additional 30 GPU hours per week.</p>\n</blockquote>",
      "rawMarkdown": "One option is to buy a Google Colab subscription for $10 USD / month. If you link your Kaggle account to your Colab account you're granted another 15 hours weekly. You also end up with the extra colab hours that you get per month with the plan. Given that's around 60 hours 2xT4 extra for the Pro plan per month for $10 it's a good deal if you're actively utilizing the hours.\n\nhttps://www.kaggle.com/blog/level-up-your-compute-more-gpu-hours-on-kaggle-wit\n\n>    Colab Pro subscribers: Get an additional 15 GPU hours per week.\n>    Colab Pro+ subscribers: Get an additional 30 GPU hours per week.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3387440,
      "author_name": "choudharymanas",
      "author_url": "",
      "post_date": "01/07/2026 02:45:54",
      "content": "<p>I always run out of my gpu quota in 3 days, But I always 2-4 hours as backup, I think you should also start doing this from next week.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3387446,
      "author_name": "madarshbb",
      "author_url": "",
      "post_date": "01/07/2026 03:04:07",
      "content": "<p>Ideally, I always have 30 mins of buffer GPU quota for a week which I plan to keep unused. If I reach close to the gpu weekly limit, I do the public LB sub first and the validation experiment later.</p>\n<p>But I try to prioritize training on TPU as much as possible but even that proves to be difficult sometimes due to the extensive queue for TPU's (sometimes 2-3 hours lol)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3387746,
      "author_name": "dragonzhang",
      "author_url": "",
      "post_date": "01/07/2026 15:20:16",
      "content": "<p>Thanks for all reply.   IT takes relatively long for training.  If I reserve submission gpu, then I can't finish my training. Anyway, resources limited.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3388034,
      "author_name": "timothylovett",
      "author_url": "",
      "post_date": "01/08/2026 02:24:54",
      "content": "<p>One option is to buy a Google Colab subscription for $10 USD / month. If you link your Kaggle account to your Colab account you're granted another 15 hours weekly. You also end up with the extra colab hours that you get per month with the plan. Given that's around 60 hours 2xT4 extra for the Pro plan per month for $10 it's a good deal if you're actively utilizing the hours.</p>\n<p><a href=\"https://www.kaggle.com/blog/level-up-your-compute-more-gpu-hours-on-kaggle-wit\" target=\"_blank\">https://www.kaggle.com/blog/level-up-your-compute-more-gpu-hours-on-kaggle-wit</a></p>\n<blockquote>\n  <p>Colab Pro subscribers: Get an additional 15 GPU hours per week.\n     Colab Pro+ subscribers: Get an additional 30 GPU hours per week.</p>\n</blockquote>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3387219": "Running out of GPU due to training,  submission is impossible anymore.\nIs it possible to provide one submission  per day?",
    "3387440": "I always run out of my gpu quota in 3 days, But I always 2-4 hours as backup, I think you should also start doing this from next week.",
    "3387446": "Ideally, I always have 30 mins of buffer GPU quota for a week which I plan to keep unused. If I reach close to the gpu weekly limit, I do the public LB sub first and the validation experiment later.\n\nBut I try to prioritize training on TPU as much as possible but even that proves to be difficult sometimes due to the extensive queue for TPU's (sometimes 2-3 hours lol)",
    "3387746": "Thanks for all reply.   IT takes relatively long for training.  If I reserve submission gpu, then I can't finish my training. Anyway, resources limited.",
    "3388034": "One option is to buy a Google Colab subscription for $10 USD / month. If you link your Kaggle account to your Colab account you're granted another 15 hours weekly. You also end up with the extra colab hours that you get per month with the plan. Given that's around 60 hours 2xT4 extra for the Pro plan per month for $10 it's a good deal if you're actively utilizing the hours.\n\nhttps://www.kaggle.com/blog/level-up-your-compute-more-gpu-hours-on-kaggle-wit\n\n>    Colab Pro subscribers: Get an additional 15 GPU hours per week.\n>    Colab Pro+ subscribers: Get an additional 30 GPU hours per week."
  },
  "source": "meta"
}