{
  "id": 189530,
  "title": "Google Cloud",
  "url": "/competitions/lyft-motion-prediction-autonomous-vehicles/discussion/189530",
  "author_name": "",
  "post_date": "2020-10-07T21:24:05.704427900Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hello everyone,</p>\n<p>I would like to move to a cloud environment and have decided to go to the Google Cloud. At a previous challenge I had already intensively used Google Colab (Pro) but was not quite convinced of its stability. I would therefore like to use a \"real\" VM in the cloud now. Does anyone have any recommendations for the hardware specs? Is a GPU really necessary? I don't have any customized rasterizers yet, so I could probably do without them, right?</p>\n<p>Thanks and best wishes</p>",
  "messages": [
    {
      "id": "1041647",
      "postDate": "10/07/2020 21:24:05",
      "content": "<p>Hello everyone,</p>\n<p>I would like to move to a cloud environment and have decided to go to the Google Cloud. At a previous challenge I had already intensively used Google Colab (Pro) but was not quite convinced of its stability. I would therefore like to use a \"real\" VM in the cloud now. Does anyone have any recommendations for the hardware specs? Is a GPU really necessary? I don't have any customized rasterizers yet, so I could probably do without them, right?</p>\n<p>Thanks and best wishes</p>",
      "rawMarkdown": "Hello everyone,\n\nI would like to move to a cloud environment and have decided to go to the Google Cloud. At a previous challenge I had already intensively used Google Colab (Pro) but was not quite convinced of its stability. I would therefore like to use a \"real\" VM in the cloud now. Does anyone have any recommendations for the hardware specs? Is a GPU really necessary? I don't have any customized rasterizers yet, so I could probably do without them, right?\n\nThanks and best wishes",
      "votes": null
    },
    {
      "id": "1041654",
      "postDate": "10/07/2020 21:29:19",
      "content": "<p><a href=\"https://www.kaggle.com/benbla\" target=\"_blank\">@benbla</a> are you using GPU or TPU in colab pro ? and the ram you are using ? tfrecords or without ?</p>",
      "rawMarkdown": "benbla are you using GPU or TPU in colab pro ? and the ram you are using ? tfrecords or without ?",
      "votes": null
    },
    {
      "id": "1043133",
      "postDate": "10/08/2020 17:57:16",
      "content": "<p><a href=\"https://www.kaggle.com/benbla\" target=\"_blank\">@benbla</a> For Google cloud, its difficult to give specific VM and GPU spec recommendations since that depends on how many training iterations you're planning to run and more importantly how much money you're willing to pay. </p>\n<p>The \"better faster\" hardware costs run quite high, so I would recommend to sticking with one of the most basic configurations (for me that ended up being 8vCPUs, 30GB RAM, GPU). For me that works ok, also with the fact that (sometimes) I have access to a local machine. So just go ahead, experiment and see what works for your situation while keeping cost in mind.</p>",
      "rawMarkdown": "benbla For Google cloud, its difficult to give specific VM and GPU spec recommendations since that depends on how many training iterations you're planning to run and more importantly how much money you're willing to pay. \n\nThe \"better faster\" hardware costs run quite high, so I would recommend to sticking with one of the most basic configurations (for me that ended up being 8vCPUs, 30GB RAM, GPU). For me that works ok, also with the fact that (sometimes) I have access to a local machine. So just go ahead, experiment and see what works for your situation while keeping cost in mind.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1041654,
      "author_name": "seshurajup",
      "author_url": "",
      "post_date": "10/07/2020 21:29:19",
      "content": "<p><a href=\"https://www.kaggle.com/benbla\" target=\"_blank\">@benbla</a> are you using GPU or TPU in colab pro ? and the ram you are using ? tfrecords or without ?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1043133,
      "author_name": "indswetrust",
      "author_url": "",
      "post_date": "10/08/2020 17:57:16",
      "content": "<p><a href=\"https://www.kaggle.com/benbla\" target=\"_blank\">@benbla</a> For Google cloud, its difficult to give specific VM and GPU spec recommendations since that depends on how many training iterations you're planning to run and more importantly how much money you're willing to pay. </p>\n<p>The \"better faster\" hardware costs run quite high, so I would recommend to sticking with one of the most basic configurations (for me that ended up being 8vCPUs, 30GB RAM, GPU). For me that works ok, also with the fact that (sometimes) I have access to a local machine. So just go ahead, experiment and see what works for your situation while keeping cost in mind.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1041647": "Hello everyone,\n\nI would like to move to a cloud environment and have decided to go to the Google Cloud. At a previous challenge I had already intensively used Google Colab (Pro) but was not quite convinced of its stability. I would therefore like to use a \"real\" VM in the cloud now. Does anyone have any recommendations for the hardware specs? Is a GPU really necessary? I don't have any customized rasterizers yet, so I could probably do without them, right?\n\nThanks and best wishes",
    "1041654": "benbla are you using GPU or TPU in colab pro ? and the ram you are using ? tfrecords or without ?",
    "1043133": "benbla For Google cloud, its difficult to give specific VM and GPU spec recommendations since that depends on how many training iterations you're planning to run and more importantly how much money you're willing to pay. \n\nThe \"better faster\" hardware costs run quite high, so I would recommend to sticking with one of the most basic configurations (for me that ended up being 8vCPUs, 30GB RAM, GPU). For me that works ok, also with the fact that (sometimes) I have access to a local machine. So just go ahead, experiment and see what works for your situation while keeping cost in mind."
  },
  "source": "meta"
}