{
  "id": 375297,
  "title": "Increase your RAM to 330GB on kaggle.",
  "url": "/competitions/otto-recommender-system/discussion/375297",
  "author_name": "",
  "post_date": "2022-12-31T14:44:23.324258200Z",
  "votes": 54,
  "comment_count": 26,
  "views": 0,
  "content": "<p>Kaggle has recently released TPU 1VM v3-8, which increases the RAM to 330 GB when activated.</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5005351%2F9862574348e463987002e861350b0500%2FScreenshot_20221231_081742.png?generation=1672498123288113&amp;alt=media\" alt=\"\"></p>\n<p>Enjoy!</p>",
  "messages": [
    {
      "id": "2081722",
      "postDate": "12/31/2022 14:44:23",
      "content": "<p>Kaggle has recently released TPU 1VM v3-8, which increases the RAM to 330 GB when activated.</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5005351%2F9862574348e463987002e861350b0500%2FScreenshot_20221231_081742.png?generation=1672498123288113&amp;alt=media\" alt=\"\"></p>\n<p>Enjoy!</p>",
      "rawMarkdown": "Kaggle has recently released TPU 1VM v3-8, which increases the RAM to 330 GB when activated.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5005351%2F9862574348e463987002e861350b0500%2FScreenshot_20221231_081742.png?generation=1672498123288113&alt=media)\n\nEnjoy!",
      "votes": null
    },
    {
      "id": "2081912",
      "postDate": "12/31/2022 20:05:57",
      "content": "<p>What is this magic?! Wow <a href=\"https://www.kaggle.com/chaudharypriyanshu\" target=\"_blank\">@chaudharypriyanshu</a> that is amazing!!!</p>\n<p>I wonder if we have the specs on how many CPU cores we have available and what the usage quota is?</p>\n<p>This can go such a long way in allowing one to preprocess large datasets with ease, without having to spend an exorbitant amount of time on minimizing the memory footprint etc. Or it can straight up allow certain operations that were not possible before even with all the memory-optimization tricks in the book.</p>\n<p>Thanks for the heads up on this! 🙌</p>",
      "rawMarkdown": "What is this magic?! Wow @chaudharypriyanshu that is amazing!!!\n\nI wonder if we have the specs on how many CPU cores we have available and what the usage quota is?\n\nThis can go such a long way in allowing one to preprocess large datasets with ease, without having to spend an exorbitant amount of time on minimizing the memory footprint etc. Or it can straight up allow certain operations that were not possible before even with all the memory-optimization tricks in the book.\n\nThanks for the heads up on this! 🙌",
      "votes": null
    },
    {
      "id": "2082054",
      "postDate": "01/01/2023 04:03:28",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/radek1\" target=\"_blank\">@radek1</a> ,<br>\nI used the following code to calculate the number of cores.</p>\n<pre><code> multiprocessing  cpu_count\n\nn_cores = cpu_count()\n\n()\n</code></pre>\n<p>There are total 96 Logical CPU cores.</p>",
      "rawMarkdown": "Hi @radek1 ,\nI used the following code to calculate the number of cores.\n```python\nfrom multiprocessing import cpu_count\n# get the number of logical cpu cores\nn_cores = cpu_count()\n# report the number of logical cpu cores\nprint(f'Number of Logical CPU cores: {n_cores}')\n```\nThere are total 96 Logical CPU cores.",
      "votes": null
    },
    {
      "id": "2082240",
      "postDate": "01/01/2023 10:24:16",
      "content": "<p>Thank you so much for the reply, <a href=\"https://www.kaggle.com/chaudharypriyanshu\" target=\"_blank\">@chaudharypriyanshu</a>! Appreciate it! 🙂 </p>",
      "rawMarkdown": "Thank you so much for the reply, @chaudharypriyanshu! Appreciate it! 🙂",
      "votes": null
    },
    {
      "id": "2082258",
      "postDate": "01/01/2023 11:02:46",
      "content": "<p>😈😈👌 Another Kaggle secret published </p>",
      "rawMarkdown": "😈😈👌 Another Kaggle secret published",
      "votes": null
    },
    {
      "id": "2082303",
      "postDate": "01/01/2023 12:27:55",
      "content": "<p>Thanks for letting us know <a href=\"https://www.kaggle.com/Nischay\" target=\"_blank\">@Nischay</a>, <strong>it was a secret</strong> 👀.</p>",
      "rawMarkdown": "Thanks for letting us know @Nischay, **it was a secret** 👀.",
      "votes": null
    },
    {
      "id": "2082363",
      "postDate": "01/01/2023 13:40:15",
      "content": "<p>Amazing !! <br>\nI was knowing about this yet <br>\nThanks a ton <a href=\"https://www.kaggle.com/chaudharypriyanshu\" target=\"_blank\">@chaudharypriyanshu</a> </p>",
      "rawMarkdown": "Amazing !! \nI was knowing about this yet \nThanks a ton @chaudharypriyanshu",
      "votes": null
    },
    {
      "id": "2082418",
      "postDate": "01/01/2023 15:03:40",
      "content": "<p>Thank you for the sharing!! </p>",
      "rawMarkdown": "Thank you for the sharing!!",
      "votes": null
    },
    {
      "id": "2082978",
      "postDate": "01/02/2023 07:05:49",
      "content": "<p>Thanks for sharing. I think there are a lot of approaches that can be used if there are enough resources like that</p>",
      "rawMarkdown": "Thanks for sharing. I think there are a lot of approaches that can be used if there are enough resources like that",
      "votes": null
    },
    {
      "id": "2082994",
      "postDate": "01/02/2023 07:17:52",
      "content": "<p>Maybe Kaggle should add CPU, IO or RAM optimized VMs to their fleet. They could be useful for different problems and they might reduce the queues on the current ones.</p>",
      "rawMarkdown": "Maybe Kaggle should add CPU, IO or RAM optimized VMs to their fleet. They could be useful for different problems and they might reduce the queues on the current ones.",
      "votes": null
    },
    {
      "id": "2083294",
      "postDate": "01/02/2023 11:50:00",
      "content": "<p>It seems that 96 GPUs and 330GB RAM come from the backend spec of TPU v3-8 instance.</p>",
      "rawMarkdown": "It seems that 96 GPUs and 330GB RAM come from the backend spec of TPU v3-8 instance.",
      "votes": null
    },
    {
      "id": "2084199",
      "postDate": "01/03/2023 09:33:51",
      "content": "<p>thank you</p>",
      "rawMarkdown": "thank you",
      "votes": null
    },
    {
      "id": "2084262",
      "postDate": "01/03/2023 11:05:46",
      "content": "<p>That is a terrific suggestion <a href=\"https://www.kaggle.com/gunesevitan\" target=\"_blank\">@gunesevitan</a>! To for instance have a chance to run with a lot of RAM for preprocessing and then a lot of CPUs for actual training… that would make such a big difference!</p>\n<p>Great suggestion! 🙌 </p>",
      "rawMarkdown": "That is a terrific suggestion @gunesevitan! To for instance have a chance to run with a lot of RAM for preprocessing and then a lot of CPUs for actual training... that would make such a big difference!\n\nGreat suggestion! 🙌",
      "votes": null
    },
    {
      "id": "2084263",
      "postDate": "01/03/2023 11:06:42",
      "content": "<p>Yup <a href=\"https://www.kaggle.com/bibanh\" target=\"_blank\">@bibanh</a>, that would make quite a difference! Many people could quite likely generate many more candidates if they were not bounded by RAM. Completely agree with your comment 🙂</p>",
      "rawMarkdown": "Yup @bibanh, that would make quite a difference! Many people could quite likely generate many more candidates if they were not bounded by RAM. Completely agree with your comment 🙂",
      "votes": null
    },
    {
      "id": "2084619",
      "postDate": "01/03/2023 16:20:17",
      "content": "<p>but how do you use the tpu? device = \"gpu\" if torch.cuda.is_available() else \"cpu\", here should someone replace gpu with tpu?</p>",
      "rawMarkdown": "but how do you use the tpu? device = \"gpu\" if torch.cuda.is_available() else \"cpu\", here should someone replace gpu with tpu?",
      "votes": null
    },
    {
      "id": "2084651",
      "postDate": "01/03/2023 16:39:43",
      "content": "<p>If you want to use TPU for RAM, then you don't need to add anything. Just turn it on and you will get the your desired 330GB.</p>",
      "rawMarkdown": "If you want to use TPU for RAM, then you don't need to add anything. Just turn it on and you will get the your desired 330GB.",
      "votes": null
    },
    {
      "id": "2084656",
      "postDate": "01/03/2023 16:45:05",
      "content": "<p>Thanks for the info</p>",
      "rawMarkdown": "Thanks for the info",
      "votes": null
    },
    {
      "id": "2085998",
      "postDate": "01/04/2023 13:56:19",
      "content": "<p>Very amazing and helpful !!! I was knowing about this yet, Thanks for your sharing <a href=\"https://www.kaggle.com/chaudharypriyanshu\" target=\"_blank\">@chaudharypriyanshu</a> </p>",
      "rawMarkdown": "Very amazing and helpful !!! I was knowing about this yet, Thanks for your sharing @chaudharypriyanshu",
      "votes": null
    },
    {
      "id": "2087183",
      "postDate": "01/05/2023 11:58:07",
      "content": "<p>I'm flying 😄</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F214989%2Fb5a3240e99831c774de5312e79840dbf%2Fflying.png?generation=1672919834462496&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "I'm flying 😄\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F214989%2Fb5a3240e99831c774de5312e79840dbf%2Fflying.png?generation=1672919834462496&alt=media)",
      "votes": null
    },
    {
      "id": "2090390",
      "postDate": "01/07/2023 09:21:47",
      "content": "<p>very helpful👍👍👍</p>",
      "rawMarkdown": "very helpful👍👍👍",
      "votes": null
    },
    {
      "id": "2093630",
      "postDate": "01/10/2023 06:53:02",
      "content": "<p>Thanks for the info.</p>",
      "rawMarkdown": "Thanks for the info.",
      "votes": null
    },
    {
      "id": "2094061",
      "postDate": "01/10/2023 15:28:36",
      "content": "<p>Wow, amazing…</p>",
      "rawMarkdown": "Wow, amazing...",
      "votes": null
    },
    {
      "id": "2096300",
      "postDate": "01/12/2023 00:22:47",
      "content": "<p>It's interesting to know that and very helpful info!! Thank you for sharing that :) 🙌</p>",
      "rawMarkdown": "It's interesting to know that and very helpful info!! Thank you for sharing that :) 🙌",
      "votes": null
    },
    {
      "id": "2096505",
      "postDate": "01/12/2023 05:14:20",
      "content": "<p>it might save my a lot money ，thanks to share</p>",
      "rawMarkdown": "it might save my a lot money ，thanks to share",
      "votes": null
    },
    {
      "id": "2097051",
      "postDate": "01/12/2023 12:38:29",
      "content": "<p>thanks for sharing</p>",
      "rawMarkdown": "thanks for sharing",
      "votes": null
    },
    {
      "id": "2097375",
      "postDate": "01/12/2023 16:13:53",
      "content": "<p>Thank you for the sharing!!！！</p>",
      "rawMarkdown": "Thank you for the sharing!!！！",
      "votes": null
    },
    {
      "id": "2120330",
      "postDate": "01/29/2023 14:13:34",
      "content": "<p>Nice, thanks!</p>",
      "rawMarkdown": "Nice, thanks!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2081912,
      "author_name": "radek1",
      "author_url": "",
      "post_date": "12/31/2022 20:05:57",
      "content": "<p>What is this magic?! Wow <a href=\"https://www.kaggle.com/chaudharypriyanshu\" target=\"_blank\">@chaudharypriyanshu</a> that is amazing!!!</p>\n<p>I wonder if we have the specs on how many CPU cores we have available and what the usage quota is?</p>\n<p>This can go such a long way in allowing one to preprocess large datasets with ease, without having to spend an exorbitant amount of time on minimizing the memory footprint etc. Or it can straight up allow certain operations that were not possible before even with all the memory-optimization tricks in the book.</p>\n<p>Thanks for the heads up on this! 🙌</p>",
      "votes": null,
      "replies": [
        {
          "id": 2082054,
          "author_name": "chaudharypriyanshu",
          "author_url": "",
          "post_date": "01/01/2023 04:03:28",
          "content": "<p>Hi <a href=\"https://www.kaggle.com/radek1\" target=\"_blank\">@radek1</a> ,<br>\nI used the following code to calculate the number of cores.</p>\n<pre><code> multiprocessing  cpu_count\n\nn_cores = cpu_count()\n\n()\n</code></pre>\n<p>There are total 96 Logical CPU cores.</p>",
          "votes": null,
          "replies": [
            {
              "id": 2082240,
              "author_name": "radek1",
              "author_url": "",
              "post_date": "01/01/2023 10:24:16",
              "content": "<p>Thank you so much for the reply, <a href=\"https://www.kaggle.com/chaudharypriyanshu\" target=\"_blank\">@chaudharypriyanshu</a>! Appreciate it! 🙂 </p>",
              "votes": null,
              "replies": []
            },
            {
              "id": 2087183,
              "author_name": "steubk",
              "author_url": "",
              "post_date": "01/05/2023 11:58:07",
              "content": "<p>I'm flying 😄</p>\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F214989%2Fb5a3240e99831c774de5312e79840dbf%2Fflying.png?generation=1672919834462496&amp;alt=media\" alt=\"\"></p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 2082258,
      "author_name": "nischaydnk",
      "author_url": "",
      "post_date": "01/01/2023 11:02:46",
      "content": "<p>😈😈👌 Another Kaggle secret published </p>",
      "votes": null,
      "replies": [
        {
          "id": 2082303,
          "author_name": "chaudharypriyanshu",
          "author_url": "",
          "post_date": "01/01/2023 12:27:55",
          "content": "<p>Thanks for letting us know <a href=\"https://www.kaggle.com/Nischay\" target=\"_blank\">@Nischay</a>, <strong>it was a secret</strong> 👀.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2082363,
      "author_name": "durgeshrao9993",
      "author_url": "",
      "post_date": "01/01/2023 13:40:15",
      "content": "<p>Amazing !! <br>\nI was knowing about this yet <br>\nThanks a ton <a href=\"https://www.kaggle.com/chaudharypriyanshu\" target=\"_blank\">@chaudharypriyanshu</a> </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2082418,
      "author_name": "eugeneryu",
      "author_url": "",
      "post_date": "01/01/2023 15:03:40",
      "content": "<p>Thank you for the sharing!! </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2082978,
      "author_name": "bibanh",
      "author_url": "",
      "post_date": "01/02/2023 07:05:49",
      "content": "<p>Thanks for sharing. I think there are a lot of approaches that can be used if there are enough resources like that</p>",
      "votes": null,
      "replies": [
        {
          "id": 2084263,
          "author_name": "radek1",
          "author_url": "",
          "post_date": "01/03/2023 11:06:42",
          "content": "<p>Yup <a href=\"https://www.kaggle.com/bibanh\" target=\"_blank\">@bibanh</a>, that would make quite a difference! Many people could quite likely generate many more candidates if they were not bounded by RAM. Completely agree with your comment 🙂</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2082994,
      "author_name": "gunesevitan",
      "author_url": "",
      "post_date": "01/02/2023 07:17:52",
      "content": "<p>Maybe Kaggle should add CPU, IO or RAM optimized VMs to their fleet. They could be useful for different problems and they might reduce the queues on the current ones.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2084262,
          "author_name": "radek1",
          "author_url": "",
          "post_date": "01/03/2023 11:05:46",
          "content": "<p>That is a terrific suggestion <a href=\"https://www.kaggle.com/gunesevitan\" target=\"_blank\">@gunesevitan</a>! To for instance have a chance to run with a lot of RAM for preprocessing and then a lot of CPUs for actual training… that would make such a big difference!</p>\n<p>Great suggestion! 🙌 </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2083294,
      "author_name": "toshik",
      "author_url": "",
      "post_date": "01/02/2023 11:50:00",
      "content": "<p>It seems that 96 GPUs and 330GB RAM come from the backend spec of TPU v3-8 instance.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2084199,
      "author_name": "kondapalkalarithvik",
      "author_url": "",
      "post_date": "01/03/2023 09:33:51",
      "content": "<p>thank you</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2084619,
      "author_name": "amitpress",
      "author_url": "",
      "post_date": "01/03/2023 16:20:17",
      "content": "<p>but how do you use the tpu? device = \"gpu\" if torch.cuda.is_available() else \"cpu\", here should someone replace gpu with tpu?</p>",
      "votes": null,
      "replies": [
        {
          "id": 2084651,
          "author_name": "chaudharypriyanshu",
          "author_url": "",
          "post_date": "01/03/2023 16:39:43",
          "content": "<p>If you want to use TPU for RAM, then you don't need to add anything. Just turn it on and you will get the your desired 330GB.</p>",
          "votes": null,
          "replies": [
            {
              "id": 2084656,
              "author_name": "amitpress",
              "author_url": "",
              "post_date": "01/03/2023 16:45:05",
              "content": "<p>Thanks for the info</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 2085998,
      "author_name": "songqizhou",
      "author_url": "",
      "post_date": "01/04/2023 13:56:19",
      "content": "<p>Very amazing and helpful !!! I was knowing about this yet, Thanks for your sharing <a href=\"https://www.kaggle.com/chaudharypriyanshu\" target=\"_blank\">@chaudharypriyanshu</a> </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2090390,
      "author_name": "ningyanggege",
      "author_url": "",
      "post_date": "01/07/2023 09:21:47",
      "content": "<p>very helpful👍👍👍</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2093630,
      "author_name": "kshigeta",
      "author_url": "",
      "post_date": "01/10/2023 06:53:02",
      "content": "<p>Thanks for the info.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2094061,
      "author_name": "primack",
      "author_url": "",
      "post_date": "01/10/2023 15:28:36",
      "content": "<p>Wow, amazing…</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2096300,
      "author_name": "yyyolanda929",
      "author_url": "",
      "post_date": "01/12/2023 00:22:47",
      "content": "<p>It's interesting to know that and very helpful info!! Thank you for sharing that :) 🙌</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2096505,
      "author_name": "xianzwaikato",
      "author_url": "",
      "post_date": "01/12/2023 05:14:20",
      "content": "<p>it might save my a lot money ，thanks to share</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2097051,
      "author_name": "soever",
      "author_url": "",
      "post_date": "01/12/2023 12:38:29",
      "content": "<p>thanks for sharing</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2097375,
      "author_name": "lonex666",
      "author_url": "",
      "post_date": "01/12/2023 16:13:53",
      "content": "<p>Thank you for the sharing!!！！</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2120330,
      "author_name": "deanpp",
      "author_url": "",
      "post_date": "01/29/2023 14:13:34",
      "content": "<p>Nice, thanks!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2081722": "Kaggle has recently released TPU 1VM v3-8, which increases the RAM to 330 GB when activated.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5005351%2F9862574348e463987002e861350b0500%2FScreenshot_20221231_081742.png?generation=1672498123288113&alt=media)\n\nEnjoy!",
    "2081912": "What is this magic?! Wow @chaudharypriyanshu that is amazing!!!\n\nI wonder if we have the specs on how many CPU cores we have available and what the usage quota is?\n\nThis can go such a long way in allowing one to preprocess large datasets with ease, without having to spend an exorbitant amount of time on minimizing the memory footprint etc. Or it can straight up allow certain operations that were not possible before even with all the memory-optimization tricks in the book.\n\nThanks for the heads up on this! 🙌",
    "2082054": "Hi @radek1 ,\nI used the following code to calculate the number of cores.\n```python\nfrom multiprocessing import cpu_count\n# get the number of logical cpu cores\nn_cores = cpu_count()\n# report the number of logical cpu cores\nprint(f'Number of Logical CPU cores: {n_cores}')\n```\nThere are total 96 Logical CPU cores.",
    "2082240": "Thank you so much for the reply, @chaudharypriyanshu! Appreciate it! 🙂",
    "2082258": "😈😈👌 Another Kaggle secret published",
    "2082303": "Thanks for letting us know @Nischay, **it was a secret** 👀.",
    "2082363": "Amazing !! \nI was knowing about this yet \nThanks a ton @chaudharypriyanshu",
    "2082418": "Thank you for the sharing!!",
    "2082978": "Thanks for sharing. I think there are a lot of approaches that can be used if there are enough resources like that",
    "2082994": "Maybe Kaggle should add CPU, IO or RAM optimized VMs to their fleet. They could be useful for different problems and they might reduce the queues on the current ones.",
    "2083294": "It seems that 96 GPUs and 330GB RAM come from the backend spec of TPU v3-8 instance.",
    "2084199": "thank you",
    "2084262": "That is a terrific suggestion @gunesevitan! To for instance have a chance to run with a lot of RAM for preprocessing and then a lot of CPUs for actual training... that would make such a big difference!\n\nGreat suggestion! 🙌",
    "2084263": "Yup @bibanh, that would make quite a difference! Many people could quite likely generate many more candidates if they were not bounded by RAM. Completely agree with your comment 🙂",
    "2084619": "but how do you use the tpu? device = \"gpu\" if torch.cuda.is_available() else \"cpu\", here should someone replace gpu with tpu?",
    "2084651": "If you want to use TPU for RAM, then you don't need to add anything. Just turn it on and you will get the your desired 330GB.",
    "2084656": "Thanks for the info",
    "2085998": "Very amazing and helpful !!! I was knowing about this yet, Thanks for your sharing @chaudharypriyanshu",
    "2087183": "I'm flying 😄\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F214989%2Fb5a3240e99831c774de5312e79840dbf%2Fflying.png?generation=1672919834462496&alt=media)",
    "2090390": "very helpful👍👍👍",
    "2093630": "Thanks for the info.",
    "2094061": "Wow, amazing...",
    "2096300": "It's interesting to know that and very helpful info!! Thank you for sharing that :) 🙌",
    "2096505": "it might save my a lot money ，thanks to share",
    "2097051": "thanks for sharing",
    "2097375": "Thank you for the sharing!!！！",
    "2120330": "Nice, thanks!"
  },
  "source": "meta"
}