{
  "id": 270510,
  "title": "Amazon EC2 Instance for the competition ",
  "url": "/competitions/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/270510",
  "author_name": "",
  "post_date": "2021-09-05T17:51:38.862828Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I am trying to select a suitable AWS EC2 instance with GPU for the competition. Any recommendation on the instance that should work with processing the data and training the model</p>",
  "messages": [
    {
      "id": "1503794",
      "postDate": "09/05/2021 17:51:38",
      "content": "<p>I am trying to select a suitable AWS EC2 instance with GPU for the competition. Any recommendation on the instance that should work with processing the data and training the model</p>",
      "rawMarkdown": "I am trying to select a suitable AWS EC2 instance with GPU for the competition. Any recommendation on the instance that should work with processing the data and training the model",
      "votes": null
    },
    {
      "id": "1503992",
      "postDate": "09/06/2021 01:00:40",
      "content": "<p>Hi Vinay, most often I would use </p>\n<ol>\n<li><strong>g4dn</strong> instance, xlarge is a good starting point, larger CPU is needed for larger models. </li>\n<li><strong>inf1</strong> instances are good for inference if you decide to split training and inference. </li>\n</ol>\n<p>A pro tip is to activate Spot instances. Spot instances are like a temporary vm, everything gets deleted once you close it. Also, it may be forced to closed if the number of regional instances run out, and your instance will be wiped an given to someone else. So why use them? You can expect to save around 70% on the price, so use it knowing the risk. </p>\n<p>Good luck!</p>",
      "rawMarkdown": "Hi Vinay, most often I would use \n1. **g4dn** instance, xlarge is a good starting point, larger CPU is needed for larger models. \n2. **inf1** instances are good for inference if you decide to split training and inference. \n\nA pro tip is to activate Spot instances. Spot instances are like a temporary vm, everything gets deleted once you close it. Also, it may be forced to closed if the number of regional instances run out, and your instance will be wiped an given to someone else. So why use them? You can expect to save around 70% on the price, so use it knowing the risk. \n\nGood luck!",
      "votes": null
    },
    {
      "id": "1504031",
      "postDate": "09/06/2021 02:53:34",
      "content": "<p>I'm working on a g4dn.xlarge instance (4 vcpu's) haven't come across any bottlenecks or issues yet</p>",
      "rawMarkdown": "I'm working on a g4dn.xlarge instance (4 vcpu's) haven't come across any bottlenecks or issues yet",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1503992,
      "author_name": "realtimshady",
      "author_url": "",
      "post_date": "09/06/2021 01:00:40",
      "content": "<p>Hi Vinay, most often I would use </p>\n<ol>\n<li><strong>g4dn</strong> instance, xlarge is a good starting point, larger CPU is needed for larger models. </li>\n<li><strong>inf1</strong> instances are good for inference if you decide to split training and inference. </li>\n</ol>\n<p>A pro tip is to activate Spot instances. Spot instances are like a temporary vm, everything gets deleted once you close it. Also, it may be forced to closed if the number of regional instances run out, and your instance will be wiped an given to someone else. So why use them? You can expect to save around 70% on the price, so use it knowing the risk. </p>\n<p>Good luck!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1504031,
      "author_name": "aryamansharma47",
      "author_url": "",
      "post_date": "09/06/2021 02:53:34",
      "content": "<p>I'm working on a g4dn.xlarge instance (4 vcpu's) haven't come across any bottlenecks or issues yet</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1503794": "I am trying to select a suitable AWS EC2 instance with GPU for the competition. Any recommendation on the instance that should work with processing the data and training the model",
    "1503992": "Hi Vinay, most often I would use \n1. **g4dn** instance, xlarge is a good starting point, larger CPU is needed for larger models. \n2. **inf1** instances are good for inference if you decide to split training and inference. \n\nA pro tip is to activate Spot instances. Spot instances are like a temporary vm, everything gets deleted once you close it. Also, it may be forced to closed if the number of regional instances run out, and your instance will be wiped an given to someone else. So why use them? You can expect to save around 70% on the price, so use it knowing the risk. \n\nGood luck!",
    "1504031": "I'm working on a g4dn.xlarge instance (4 vcpu's) haven't come across any bottlenecks or issues yet"
  },
  "source": "meta"
}