{ "id": 341988, "title": "What hardware are you using for training?", "url": "/competitions/AI4Code/discussion/341988", "author_name": "The Devastator", "post_date": "2022-08-05T06:15:55.852000", "votes": 13, "comment_count": 21, "views": 0, "content": "

Hi Guys,
\nHow is the final week going for you?

\n

What hardware are you using?
\nI am currently training on a local machine with 2x Titan RTX.

\n

Just asking out of curiosity to know where I stand on this topic.
\nThe competition's remaining time is starting to get tight and I got more experiments I planned than I would be able to actually run..

", "messages": [ { "id": 1885367, "postDate": "2022-08-05T06:15:55.853Z", "content": "

Hi Guys,
\nHow is the final week going for you?

\n

What hardware are you using?
\nI am currently training on a local machine with 2x Titan RTX.

\n

Just asking out of curiosity to know where I stand on this topic.
\nThe competition's remaining time is starting to get tight and I got more experiments I planned than I would be able to actually run..

", "rawMarkdown": "Hi Guys,\nHow is the final week going for you? \n\nWhat hardware are you using?\nI am currently training on a local machine with 2x Titan RTX. \n\nJust asking out of curiosity to know where I stand on this topic.\nThe competition's remaining time is starting to get tight and I got more experiments I planned than I would be able to actually run.. \n\n\n\n\n", "votes": 13 }, { "id": 1886189, "postDate": "2022-08-05T16:19:48.777Z", "content": "

I have Colab Pro+ and most of the time I get a P100, but sometimes it gives me a V100 which is around 3 times faster for me.

", "rawMarkdown": "I have Colab Pro+ and most of the time I get a P100, but sometimes it gives me a V100 which is around 3 times faster for me. ", "votes": 6, "replies": [ { "id": 1886936, "postDate": "2022-08-06T09:35:40.410Z", "content": "

Wow, your models must be really good and light weight. Hope you do share your approach post-comp ends… but maybe you have already given a hint…

", "rawMarkdown": "Wow, your models must be really good and light weight. Hope you do share your approach post-comp ends... but maybe you have already given a hint..." }, { "id": 1887358, "postDate": "2022-08-06T15:54:05.110Z", "content": "

wow, with v100 and you got so high lb. I got a100 for colab pro+ but still stuck at this lb. Hope you do share your approach after this competition ends.

", "rawMarkdown": "wow, with v100 and you got so high lb. I got a100 for colab pro+ but still stuck at this lb. Hope you do share your approach after this competition ends.", "votes": 1 }, { "id": 1892453, "postDate": "2022-08-10T05:26:05.247Z", "content": "

Hi, When I used colab pro with data in google drive the IO operations were so slow I couldn't use it for training. Did you face this issue?
\nThank you

", "rawMarkdown": "Hi, When I used colab pro with data in google drive the IO operations were so slow I couldn't use it for training. Did you face this issue? \nThank you" } ] }, { "id": 1886963, "postDate": "2022-08-06T10:11:52.410Z", "content": "

\"\"

\n

I have rent Nvidia A100 card. It is my first kaggle competition with learning big models

", "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F9632811%2Ff386853ae89030999cfd0bc32d0a3928%2FScreenshot%202022-08-06%20at%2013.10.21.png?generation=1659780633473187&alt=media)\n\nI have rent Nvidia A100 card. It is my first kaggle competition with learning big models", "votes": 3, "replies": [ { "id": 1889147, "postDate": "2022-08-08T02:36:54.787Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1889498, "postDate": "2022-08-08T08:27:31.127Z", "content": "

This looks beautiful!

", "rawMarkdown": "This looks beautiful!", "votes": 1 }, { "id": 1891409, "postDate": "2022-08-09T12:04:25.770Z", "content": "

Nice!
\nMay I ask if this is a physical device or using a cloud provider? If so, do you have any recommendations?

", "rawMarkdown": "Nice!\nMay I ask if this is a physical device or using a cloud provider? If so, do you have any recommendations?" } ] }, { "id": 1886132, "postDate": "2022-08-05T15:50:58.197Z", "content": "

🥔

", "rawMarkdown": "# 🥔 ", "votes": 3, "replies": [ { "id": 1886176, "postDate": "2022-08-05T16:16:06.547Z", "content": "

😂 ………

", "rawMarkdown": "😂 .........", "votes": 1 } ] }, { "id": 1885779, "postDate": "2022-08-05T11:25:28.337Z", "content": "

Hi friend, at least you have not only one but two GPUs… I only have 16GB RAM, imagine my frustration

", "rawMarkdown": "Hi friend, at least you have not only one but two GPUs... I only have 16GB RAM, imagine my frustration", "votes": 1, "replies": [ { "id": 1886590, "postDate": "2022-08-06T02:35:37.567Z", "content": "

You should get a medal regardless the competition results.

", "rawMarkdown": "You should get a medal regardless the competition results." }, { "id": 1887361, "postDate": "2022-08-06T16:01:17.943Z", "content": "

ha ha rofl, but seriously, I think 84 is possible with Kaggle resources (P100 I believe and around 5 hours training…. I am not fully there but confident due to CV

", "rawMarkdown": "ha ha rofl, but seriously, I think 84 is possible with Kaggle resources (P100 I believe and around 5 hours training.... I am not fully there but confident due to CV", "votes": 1 } ] }, { "id": 1886102, "postDate": "2022-08-05T15:24:56.917Z", "content": "

A 3060 12GB (about 40% of a V100), and Colab Pro+. This is my first time really doing this style of competition, so I am learning a good bit even if my models aren't :)

", "rawMarkdown": "A 3060 12GB (about 40% of a V100), and Colab Pro+. This is my first time really doing this style of competition, so I am learning a good bit even if my models aren't :)", "votes": 2, "replies": [ { "id": 1886179, "postDate": "2022-08-05T16:16:43.237Z", "content": "

That's the important thing! Good luck in the competition.

", "rawMarkdown": "That's the important thing! Good luck in the competition.", "votes": 1 }, { "id": 1888415, "postDate": "2022-08-07T14:43:38.133Z", "content": "

Thanks!

\n

It's really easy to introduce subtle bugs w/deep learning things that don't cause obvious errors, but definitely hurt results. OTOH, it led me onto a possibly interesting path, and if DataMan hadn't already confirmed it I was going to say something like \"I think you can get well into at least the medal range with one V100/3090 and the right model… and that ensemble notebook really didn't mess things up at all in the big picture\" ;)

", "rawMarkdown": "Thanks! \n\nIt's really easy to introduce subtle bugs w/deep learning things that don't cause obvious errors, but definitely hurt results. OTOH, it led me onto a possibly interesting path, and if DataMan hadn't already confirmed it I was going to say something like \"I think you can get well into at least the medal range with one V100/3090 and the right model... and that ensemble notebook really didn't mess things up *at all* in the big picture\" ;)" } ] }, { "id": 1885766, "postDate": "2022-08-05T11:20:54.167Z", "content": "

RTX 4*3090, and it was still slow.

", "rawMarkdown": "RTX 4*3090, and it was still slow.", "votes": 2 }, { "id": 1885460, "postDate": "2022-08-05T07:28:46.210Z", "content": "

At the moment I am using 2x RTX 3090. Soon there will be more :)

", "rawMarkdown": "At the moment I am using 2x RTX 3090. Soon there will be more :)", "votes": 2 }, { "id": 1894695, "postDate": "2022-08-11T16:49:19.103Z", "content": "

A5000 with each training epoch 6 hours.

", "rawMarkdown": "A5000 with each training epoch 6 hours.\n" }, { "id": 1887290, "postDate": "2022-08-06T15:25:49.203Z", "content": "

quadro rtx a6000, 24gb

", "rawMarkdown": "quadro rtx a6000, 24gb" }, { "id": 1887083, "postDate": "2022-08-06T12:25:26.157Z", "content": "

A rtx 2060 6G. It's really slow

", "rawMarkdown": "A rtx 2060 6G. It's really slow" } ], "comments": [ { "id": 1886189, "author_name": "Alireza", "author_url": "", "post_date": "2022-08-05T16:19:48.777000", "content": "

I have Colab Pro+ and most of the time I get a P100, but sometimes it gives me a V100 which is around 3 times faster for me.

", "votes": 6, "replies": [ { "id": 1886936, "author_name": "Allohvk", "author_url": "", "post_date": "2022-08-06T09:35:40.410000", "content": "

Wow, your models must be really good and light weight. Hope you do share your approach post-comp ends… but maybe you have already given a hint…

", "votes": 0, "replies": [] }, { "id": 1887358, "author_name": "Bao Loc Pham", "author_url": "", "post_date": "2022-08-06T15:54:05.110000", "content": "

wow, with v100 and you got so high lb. I got a100 for colab pro+ but still stuck at this lb. Hope you do share your approach after this competition ends.

", "votes": 1, "replies": [] }, { "id": 1892453, "author_name": "AbhilashVJ", "author_url": "", "post_date": "2022-08-10T05:26:05.247000", "content": "

Hi, When I used colab pro with data in google drive the IO operations were so slow I couldn't use it for training. Did you face this issue?
\nThank you

", "votes": 0, "replies": [] } ] }, { "id": 1886963, "author_name": "toomuch", "author_url": "", "post_date": "2022-08-06T10:11:52.410000", "content": "

\"\"

\n

I have rent Nvidia A100 card. It is my first kaggle competition with learning big models

", "votes": 3, "replies": [ { "id": 1889147, "author_name": "", "author_url": "", "post_date": "2022-08-08T02:36:54.787000", "content": "", "votes": 0, "replies": [] }, { "id": 1889498, "author_name": "CroDoc", "author_url": "", "post_date": "2022-08-08T08:27:31.127000", "content": "

This looks beautiful!

", "votes": 1, "replies": [] }, { "id": 1891409, "author_name": "The Devastator", "author_url": "", "post_date": "2022-08-09T12:04:25.770000", "content": "

Nice!
\nMay I ask if this is a physical device or using a cloud provider? If so, do you have any recommendations?

", "votes": 0, "replies": [] } ] }, { "id": 1886132, "author_name": "Nicholas Broad", "author_url": "", "post_date": "2022-08-05T15:50:58.197000", "content": "

🥔

", "votes": 3, "replies": [ { "id": 1886176, "author_name": "Bella Pisani", "author_url": "", "post_date": "2022-08-05T16:16:06.547000", "content": "

😂 ………

", "votes": 1, "replies": [] } ] }, { "id": 1885779, "author_name": "Gabriel Soares Arantes", "author_url": "", "post_date": "2022-08-05T11:25:28.337000", "content": "

Hi friend, at least you have not only one but two GPUs… I only have 16GB RAM, imagine my frustration

", "votes": 1, "replies": [ { "id": 1886590, "author_name": "Johnny Wise", "author_url": "", "post_date": "2022-08-06T02:35:37.567000", "content": "

You should get a medal regardless the competition results.

", "votes": 0, "replies": [] }, { "id": 1887361, "author_name": "Allohvk", "author_url": "", "post_date": "2022-08-06T16:01:17.943000", "content": "

ha ha rofl, but seriously, I think 84 is possible with Kaggle resources (P100 I believe and around 5 hours training…. I am not fully there but confident due to CV

", "votes": 1, "replies": [] } ] }, { "id": 1886102, "author_name": "happycube", "author_url": "", "post_date": "2022-08-05T15:24:56.917000", "content": "

A 3060 12GB (about 40% of a V100), and Colab Pro+. This is my first time really doing this style of competition, so I am learning a good bit even if my models aren't :)

", "votes": 2, "replies": [ { "id": 1886179, "author_name": "Bella Pisani", "author_url": "", "post_date": "2022-08-05T16:16:43.237000", "content": "

That's the important thing! Good luck in the competition.

", "votes": 1, "replies": [] }, { "id": 1888415, "author_name": "happycube", "author_url": "", "post_date": "2022-08-07T14:43:38.133000", "content": "

Thanks!

\n

It's really easy to introduce subtle bugs w/deep learning things that don't cause obvious errors, but definitely hurt results. OTOH, it led me onto a possibly interesting path, and if DataMan hadn't already confirmed it I was going to say something like \"I think you can get well into at least the medal range with one V100/3090 and the right model… and that ensemble notebook really didn't mess things up at all in the big picture\" ;)

", "votes": 0, "replies": [] } ] }, { "id": 1885766, "author_name": "Turbo", "author_url": "", "post_date": "2022-08-05T11:20:54.167000", "content": "

RTX 4*3090, and it was still slow.

", "votes": 2, "replies": [] }, { "id": 1885460, "author_name": "CroDoc", "author_url": "", "post_date": "2022-08-05T07:28:46.210000", "content": "

At the moment I am using 2x RTX 3090. Soon there will be more :)

", "votes": 2, "replies": [] }, { "id": 1894695, "author_name": "Wenxuan Ye", "author_url": "", "post_date": "2022-08-11T16:49:19.103000", "content": "

A5000 with each training epoch 6 hours.

", "votes": 0, "replies": [] }, { "id": 1887290, "author_name": "Mark Baushenko", "author_url": "", "post_date": "2022-08-06T15:25:49.203000", "content": "

quadro rtx a6000, 24gb

", "votes": 0, "replies": [] }, { "id": 1887083, "author_name": "Sáng Huỳnh", "author_url": "", "post_date": "2022-08-06T12:25:26.157000", "content": "

A rtx 2060 6G. It's really slow

", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1885367": "Hi Guys,\nHow is the final week going for you? \n\nWhat hardware are you using?\nI am currently training on a local machine with 2x Titan RTX. \n\nJust asking out of curiosity to know where I stand on this topic.\nThe competition's remaining time is starting to get tight and I got more experiments I planned than I would be able to actually run.. \n\n\n\n\n", "1886189": "I have Colab Pro+ and most of the time I get a P100, but sometimes it gives me a V100 which is around 3 times faster for me. ", "1886963": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F9632811%2Ff386853ae89030999cfd0bc32d0a3928%2FScreenshot%202022-08-06%20at%2013.10.21.png?generation=1659780633473187&alt=media)\n\nI have rent Nvidia A100 card. It is my first kaggle competition with learning big models", "1886132": "# 🥔 ", "1885779": "Hi friend, at least you have not only one but two GPUs... I only have 16GB RAM, imagine my frustration", "1886102": "A 3060 12GB (about 40% of a V100), and Colab Pro+. This is my first time really doing this style of competition, so I am learning a good bit even if my models aren't :)", "1885766": "RTX 4*3090, and it was still slow.", "1885460": "At the moment I am using 2x RTX 3090. Soon there will be more :)", "1894695": "A5000 with each training epoch 6 hours.\n", "1887290": "quadro rtx a6000, 24gb", "1887083": "A rtx 2060 6G. It's really slow" } }