{ "id": 238098, "title": "Hardware used for this competition?", "url": "/competitions/hpa-single-cell-image-classification/discussion/238098", "author_name": "CroDoc", "post_date": "2021-05-11T07:59:15.771000", "votes": 8, "comment_count": 30, "views": 0, "content": "

What are you guys using? What's a standard/prefered setup for grandmasters?

\n

I have a gaming PC with a RTX 3090, i7-10700KF and 64GB RAM. Wasn't easy to find just the GPU alone 2-3 motnhs ago so I just bought it together. I did buy 3 extra SSD-s after for this contest … never enough space :D

\n

I would love to hear suggestions/preferences on setups for computer vision contests…

", "messages": [ { "id": 1301719, "postDate": "2021-05-11T07:59:15.773Z", "content": "

What are you guys using? What's a standard/prefered setup for grandmasters?

\n

I have a gaming PC with a RTX 3090, i7-10700KF and 64GB RAM. Wasn't easy to find just the GPU alone 2-3 motnhs ago so I just bought it together. I did buy 3 extra SSD-s after for this contest … never enough space :D

\n

I would love to hear suggestions/preferences on setups for computer vision contests…

", "rawMarkdown": "What are you guys using? What's a standard/prefered setup for grandmasters?\n\nI have a gaming PC with a RTX 3090, i7-10700KF and 64GB RAM. Wasn't easy to find just the GPU alone 2-3 motnhs ago so I just bought it together. I did buy 3 extra SSD-s after for this contest ... never enough space :D\n\nI would love to hear suggestions/preferences on setups for computer vision contests...", "votes": 8 }, { "id": 1301776, "postDate": "2021-05-11T08:36:42.580Z", "content": "

I used Kaggle TPU

", "rawMarkdown": "I used Kaggle TPU", "votes": 8 }, { "id": 1301728, "postDate": "2021-05-11T08:02:51.027Z", "content": "

Colab Pro, 10 USD/Month for V100 GPU :)

", "rawMarkdown": "Colab Pro, 10 USD/Month for V100 GPU :)", "votes": 5, "replies": [ { "id": 1301746, "postDate": "2021-05-11T08:12:02.540Z", "content": "

I got it for 9.99 USD/month :)

", "rawMarkdown": "I got it for 9.99 USD/month :)", "votes": 1 }, { "id": 1301755, "postDate": "2021-05-11T08:15:51.730Z", "content": "

@alexanderriedel
\nI am surprised that Colab Pro never complain about how much I use it.
\ni am using TPU a lot.

", "rawMarkdown": "@alexanderriedel \nI am surprised that Colab Pro never complain about how much I use it.\ni am using TPU a lot." }, { "id": 1301759, "postDate": "2021-05-11T08:22:16.477Z", "content": "

For the GPU session I had some GPU quota exceedings, but very rarely and only after real extensive use 👍

", "rawMarkdown": "For the GPU session I had some GPU quota exceedings, but very rarely and only after real extensive use 👍" }, { "id": 1301807, "postDate": "2021-05-11T08:51:13.727Z", "content": "

Colab pro actually sounds good. But there is a 24h reset limit? Is that problematic?

", "rawMarkdown": "Colab pro actually sounds good. But there is a 24h reset limit? Is that problematic?" }, { "id": 1301816, "postDate": "2021-05-11T08:56:57.893Z", "content": "

Yes this is definately a problem but you can overcome it easily with a good model saving and resume training strategy

", "rawMarkdown": "Yes this is definately a problem but you can overcome it easily with a good model saving and resume training strategy", "votes": 3 }, { "id": 1301839, "postDate": "2021-05-11T09:08:55.400Z", "content": "

Hi @crodoc

\n

I like the most about Colab pro:

\n

1- You can use Kaggle datasets via using the google cloud storage (GCS) bucket link without downloading it and for free. since all Kaggle dataset in GCS.

\n

2- with changing the working directory in Colab and using training callbacks you can keep saving your models in your Google Drive.

\n

3- Using google cloud/Colab is a skill I need to master because from next year I will start using tensorflow cloud carefully, I feel that it needed to be Kaggle competition grandmaster.

\n

What is TensorFlow Cloud?

\n
\n

TensorFlow Cloud is a Python package that provides APIs for a seamless transition from local debugging to distributed training in Google Cloud. It simplifies the process of training TensorFlow models on the cloud into a single, simple function call, requiring minimal setup and no changes to your model. TensorFlow Cloud handles cloud-specific tasks such as creating VM instances and distribution strategies for your models automatically.

\n
\n

https://www.tensorflow.org/guide/keras/training_keras_models_on_cloud

", "rawMarkdown": "Hi @crodoc \n\nI like the most about Colab pro:\n\n 1- You can use Kaggle datasets via using the google cloud storage (GCS) bucket link without downloading it and for free. since all Kaggle dataset in GCS.\n\n2- with changing the working directory in Colab and using training callbacks you can keep saving your models in your Google Drive. \n\n3- Using google cloud/Colab is a skill I need to master because from next year I will start using tensorflow cloud carefully, I feel that it needed to be Kaggle competition grandmaster.\n\nWhat is TensorFlow Cloud?\n> TensorFlow Cloud is a Python package that provides APIs for a seamless transition from local debugging to **distributed training** in Google Cloud. It simplifies the process of training TensorFlow models on the cloud into a single, simple function call, requiring minimal setup and no changes to your model. TensorFlow Cloud handles cloud-specific tasks such as creating VM instances and distribution strategies for your models automatically.\n\nhttps://www.tensorflow.org/guide/keras/training_keras_models_on_cloud", "votes": 1 }, { "id": 1301928, "postDate": "2021-05-11T10:13:23.090Z", "content": "

I wish I checked this out 2 weeks ago… such a shame I didn't post this discussion before

", "rawMarkdown": "I wish I checked this out 2 weeks ago... such a shame I didn't post this discussion before" }, { "id": 1301933, "postDate": "2021-05-11T10:17:00.230Z", "content": "

A point is, that your 3090 has 24GB memory, the Colab GPUs only have 16GB, so I think it's not really a waste of money or something

", "rawMarkdown": "A point is, that your 3090 has 24GB memory, the Colab GPUs only have 16GB, so I think it's not really a waste of money or something" }, { "id": 1301965, "postDate": "2021-05-11T10:37:27.037Z", "content": "

@crodoc @alexanderriedel

\n

You are welcome to team up

\n

https://www.kaggle.com/c/birdclef-2021

\n

To have fun with Audio deep learning and learn from each other how to use Google colab well.

\n

I am looking forward..

", "rawMarkdown": "@crodoc @alexanderriedel \n\nYou are welcome to team up\n\nhttps://www.kaggle.com/c/birdclef-2021\n\nTo have fun with Audio deep learning and learn from each other how to use Google colab well.\n\nI am looking forward..", "votes": 1 }, { "id": 1302269, "postDate": "2021-05-11T13:13:01.330Z", "content": "

Thanks for the invite @faisalalsrheed.
\nI have not decided my next steps … still pretty tired from this contest … need sleep fast :)

", "rawMarkdown": "Thanks for the invite @faisalalsrheed.\nI have not decided my next steps ... still pretty tired from this contest ... need sleep fast :)", "votes": 2 }, { "id": 1302281, "postDate": "2021-05-11T13:20:51.113Z", "content": "

@crodoc reminded me of my first Kaggle competition :)
\nhttps://www.kaggle.com/c/jane-street-market-prediction

\n

after 4 competitions, you will start to see the fun side of Kaggle.

", "rawMarkdown": "@crodoc reminded me of my first Kaggle competition :)\nhttps://www.kaggle.com/c/jane-street-market-prediction \n\nafter 4 competitions, you will start to see the fun side of Kaggle. " } ] }, { "id": 1301828, "postDate": "2021-05-11T09:05:17.157Z", "content": "

HP laptop, Kaggle GPU, Colab Pro

", "rawMarkdown": "HP laptop, Kaggle GPU, Colab Pro", "votes": 3 }, { "id": 1301748, "postDate": "2021-05-11T08:12:56.363Z", "content": "

Colab Pro (TPU ) and Kaggle (TPU)

", "rawMarkdown": "Colab Pro (TPU ) and Kaggle (TPU) ", "votes": 3 }, { "id": 1302726, "postDate": "2021-05-11T17:20:06.810Z", "content": "

Home RTX 3090/64GB RAM, 1TB SSD.

", "rawMarkdown": "Home RTX 3090/64GB RAM, 1TB SSD.", "votes": 4, "replies": [ { "id": 1302796, "postDate": "2021-05-11T18:02:54.990Z", "content": "

I have 1+2+2+4 TB of SSD…. just because of this contest got 3 extra 😅

", "rawMarkdown": "I have 1+2+2+4 TB of SSD.... just because of this contest got 3 extra 😅", "votes": 3 } ] }, { "id": 1301871, "postDate": "2021-05-11T09:27:13.523Z", "content": "

my team member @joven1997 has 16 x V100 GPUs in the DGX box. So i used it

", "rawMarkdown": "my team member @joven1997 has 16 x V100 GPUs in the DGX box. So i used it", "votes": 2, "replies": [ { "id": 1301924, "postDate": "2021-05-11T10:11:39.203Z", "content": "

amazing :) I wish I could get my hands on one

", "rawMarkdown": "amazing :) I wish I could get my hands on one", "votes": 1 }, { "id": 1302423, "postDate": "2021-05-11T14:38:08.880Z", "content": "

@crodoc you have my e-mail :)

", "rawMarkdown": "@crodoc you have my e-mail :)" }, { "id": 1302960, "postDate": "2021-05-11T20:01:20.170Z", "content": "
\n

my team member @joven1997 has 16 x V100 GPUs in the DGX box. So i used it

\n
\n

Wow - that's pretty sweet @morizin - were you guys able to utilise all that GPU compute? How many models were you able to run concurrently and how did you keep track of all your results? Good luck with the last few hours.

", "rawMarkdown": "> my team member @joven1997 has 16 x V100 GPUs in the DGX box. So i used it\n\nWow - that's pretty sweet @morizin - were you guys able to utilise all that GPU compute? How many models were you able to run concurrently and how did you keep track of all your results? Good luck with the last few hours." }, { "id": 1302976, "postDate": "2021-05-11T20:19:58.533Z", "content": "

we didnt use all of them at a time. we used only 6 to 8 of them. We were training fold networks. but sad, this 16 x V100 GPU cannot be used in kaggle submission.

", "rawMarkdown": "we didnt use all of them at a time. we used only 6 to 8 of them. We were training fold networks. but sad, this 16 x V100 GPU cannot be used in kaggle submission." }, { "id": 1302991, "postDate": "2021-05-11T20:35:55.757Z", "content": "

@datahobbit Most of the models of our team were trained using Kaggle TPU (Keras+TF), models trained on DGX were Pytorch based, the number of models runned concurently usually were 2-3. We used https://wandb.ai/ to track results. So basically DGX was not involved much. But in general the number of models running in the same time depends on the image size, batch size and model size.

", "rawMarkdown": "@datahobbit Most of the models of our team were trained using Kaggle TPU (Keras+TF), models trained on DGX were Pytorch based, the number of models runned concurently usually were 2-3. We used https://wandb.ai/ to track results. So basically DGX was not involved much. But in general the number of models running in the same time depends on the image size, batch size and model size. ", "votes": 1 }, { "id": 1305647, "postDate": "2021-05-13T12:08:25.263Z", "content": "

@morizin and @joven1997 thank you for the explanation and congratulations on your strong result. Too bad it wasn't quite gold, I know how disappointing it can be to get so close and miss out…
\nIn a lot of competitions, I have been pretty maxed out on the hardware I had. I also know that even with much less hardware existing results and planning what to run next needed some careful thought. That's why I have wondered if I had a lot more hardware available, would I actually be able to use all of it? Your answers gave some interesting insights into that. Good luck in future competitions!

", "rawMarkdown": "@morizin and @joven1997 thank you for the explanation and congratulations on your strong result. Too bad it wasn't quite gold, I know how disappointing it can be to get so close and miss out... \nIn a lot of competitions, I have been pretty maxed out on the hardware I had. I also know that even with much less hardware existing results and planning what to run next needed some careful thought. That's why I have wondered if I had a lot more hardware available, would I actually be able to use all of it? Your answers gave some interesting insights into that. Good luck in future competitions!", "votes": 1 } ] }, { "id": 1303780, "postDate": "2021-05-12T08:32:10.610Z", "content": "

Kaggle / Colab, i m waiting for availability of RTX 3090 to build workstation

", "rawMarkdown": "Kaggle / Colab, i m waiting for availability of RTX 3090 to build workstation ", "votes": 1 }, { "id": 1303500, "postDate": "2021-05-12T05:14:41.643Z", "content": "

GeForce 970 GTX/24GB RAM/simple HDD for single cell classification model based on EfficientNetV2. It trained about 9 days with some simple optimization tricks.

", "rawMarkdown": "GeForce 970 GTX/24GB RAM/simple HDD for single cell classification model based on EfficientNetV2. It trained about 9 days with some simple optimization tricks.", "votes": 1 }, { "id": 1301832, "postDate": "2021-05-11T09:06:44.130Z", "content": "

2x GTX 1080 Ti, i7-8700K, 32 GB RAM (with a 32 GB swap 😅)
\noccasionally I used RTX 2080 Ti on another similar gaming/DL PC

", "rawMarkdown": "2x GTX 1080 Ti, i7-8700K, 32 GB RAM (with a 32 GB swap 😅)\noccasionally I used RTX 2080 Ti on another similar gaming/DL PC", "votes": 1 }, { "id": 1303519, "postDate": "2021-05-12T05:38:20.220Z", "content": "

Home GTX1080/16GB RAM + 1TB SSD + extensive use of Colab Pro :-)

", "rawMarkdown": "Home GTX1080/16GB RAM + 1TB SSD + extensive use of Colab Pro :-)", "votes": 2 }, { "id": 1302758, "postDate": "2021-05-11T17:35:42.763Z", "content": "

My sincere suggestion for you: set up your pc in a way that it reaches a good balance of brainstorming speed and experiment speed. It’s also important to take how much time you can devote to competitions. In fact, I think your setup should be able to play in most of the competitions.

", "rawMarkdown": "My sincere suggestion for you: set up your pc in a way that it reaches a good balance of brainstorming speed and experiment speed. It’s also important to take how much time you can devote to competitions. In fact, I think your setup should be able to play in most of the competitions.", "votes": 2 }, { "id": 1301865, "postDate": "2021-05-11T09:26:11.740Z", "content": "

Kaggle Notebook and Datasets (TPU training & GPU inference)

", "rawMarkdown": "Kaggle Notebook and Datasets (TPU training & GPU inference)", "votes": 2 } ], "comments": [ { "id": 1301776, "author_name": "LucaMTB", "author_url": "", "post_date": "2021-05-11T08:36:42.580000", "content": "

I used Kaggle TPU

", "votes": 8, "replies": [] }, { "id": 1301728, "author_name": "Alexander Riedel", "author_url": "", "post_date": "2021-05-11T08:02:51.027000", "content": "

Colab Pro, 10 USD/Month for V100 GPU :)

", "votes": 5, "replies": [ { "id": 1301746, "author_name": "Faisal Alsrheed", "author_url": "", "post_date": "2021-05-11T08:12:02.540000", "content": "

I got it for 9.99 USD/month :)

", "votes": 1, "replies": [] }, { "id": 1301755, "author_name": "Faisal Alsrheed", "author_url": "", "post_date": "2021-05-11T08:15:51.730000", "content": "

@alexanderriedel
\nI am surprised that Colab Pro never complain about how much I use it.
\ni am using TPU a lot.

", "votes": 0, "replies": [] }, { "id": 1301759, "author_name": "Alexander Riedel", "author_url": "", "post_date": "2021-05-11T08:22:16.477000", "content": "

For the GPU session I had some GPU quota exceedings, but very rarely and only after real extensive use 👍

", "votes": 0, "replies": [] }, { "id": 1301807, "author_name": "CroDoc", "author_url": "", "post_date": "2021-05-11T08:51:13.727000", "content": "

Colab pro actually sounds good. But there is a 24h reset limit? Is that problematic?

", "votes": 0, "replies": [] }, { "id": 1301816, "author_name": "Alexander Riedel", "author_url": "", "post_date": "2021-05-11T08:56:57.893000", "content": "

Yes this is definately a problem but you can overcome it easily with a good model saving and resume training strategy

", "votes": 3, "replies": [] }, { "id": 1301839, "author_name": "Faisal Alsrheed", "author_url": "", "post_date": "2021-05-11T09:08:55.400000", "content": "

Hi @crodoc

\n

I like the most about Colab pro:

\n

1- You can use Kaggle datasets via using the google cloud storage (GCS) bucket link without downloading it and for free. since all Kaggle dataset in GCS.

\n

2- with changing the working directory in Colab and using training callbacks you can keep saving your models in your Google Drive.

\n

3- Using google cloud/Colab is a skill I need to master because from next year I will start using tensorflow cloud carefully, I feel that it needed to be Kaggle competition grandmaster.

\n

What is TensorFlow Cloud?

\n
\n

TensorFlow Cloud is a Python package that provides APIs for a seamless transition from local debugging to distributed training in Google Cloud. It simplifies the process of training TensorFlow models on the cloud into a single, simple function call, requiring minimal setup and no changes to your model. TensorFlow Cloud handles cloud-specific tasks such as creating VM instances and distribution strategies for your models automatically.

\n
\n

https://www.tensorflow.org/guide/keras/training_keras_models_on_cloud

", "votes": 1, "replies": [] }, { "id": 1301928, "author_name": "CroDoc", "author_url": "", "post_date": "2021-05-11T10:13:23.090000", "content": "

I wish I checked this out 2 weeks ago… such a shame I didn't post this discussion before

", "votes": 0, "replies": [] }, { "id": 1301933, "author_name": "Alexander Riedel", "author_url": "", "post_date": "2021-05-11T10:17:00.230000", "content": "

A point is, that your 3090 has 24GB memory, the Colab GPUs only have 16GB, so I think it's not really a waste of money or something

", "votes": 0, "replies": [] }, { "id": 1301965, "author_name": "Faisal Alsrheed", "author_url": "", "post_date": "2021-05-11T10:37:27.037000", "content": "

@crodoc @alexanderriedel

\n

You are welcome to team up

\n

https://www.kaggle.com/c/birdclef-2021

\n

To have fun with Audio deep learning and learn from each other how to use Google colab well.

\n

I am looking forward..

", "votes": 1, "replies": [] }, { "id": 1302269, "author_name": "CroDoc", "author_url": "", "post_date": "2021-05-11T13:13:01.330000", "content": "

Thanks for the invite @faisalalsrheed.
\nI have not decided my next steps … still pretty tired from this contest … need sleep fast :)

", "votes": 2, "replies": [] }, { "id": 1302281, "author_name": "Faisal Alsrheed", "author_url": "", "post_date": "2021-05-11T13:20:51.113000", "content": "

@crodoc reminded me of my first Kaggle competition :)
\nhttps://www.kaggle.com/c/jane-street-market-prediction

\n

after 4 competitions, you will start to see the fun side of Kaggle.

", "votes": 0, "replies": [] } ] }, { "id": 1301828, "author_name": "novice03", "author_url": "", "post_date": "2021-05-11T09:05:17.157000", "content": "

HP laptop, Kaggle GPU, Colab Pro

", "votes": 3, "replies": [] }, { "id": 1301748, "author_name": "Faisal Alsrheed", "author_url": "", "post_date": "2021-05-11T08:12:56.363000", "content": "

Colab Pro (TPU ) and Kaggle (TPU)

", "votes": 3, "replies": [] }, { "id": 1302726, "author_name": "MPWARE", "author_url": "", "post_date": "2021-05-11T17:20:06.810000", "content": "

Home RTX 3090/64GB RAM, 1TB SSD.

", "votes": 4, "replies": [ { "id": 1302796, "author_name": "CroDoc", "author_url": "", "post_date": "2021-05-11T18:02:54.990000", "content": "

I have 1+2+2+4 TB of SSD…. just because of this contest got 3 extra 😅

", "votes": 3, "replies": [] } ] }, { "id": 1301871, "author_name": "Mohammed Rizin V K", "author_url": "", "post_date": "2021-05-11T09:27:13.523000", "content": "

my team member @joven1997 has 16 x V100 GPUs in the DGX box. So i used it

", "votes": 2, "replies": [ { "id": 1301924, "author_name": "CroDoc", "author_url": "", "post_date": "2021-05-11T10:11:39.203000", "content": "

amazing :) I wish I could get my hands on one

", "votes": 1, "replies": [] }, { "id": 1302423, "author_name": "wayfarer", "author_url": "", "post_date": "2021-05-11T14:38:08.880000", "content": "

@crodoc you have my e-mail :)

", "votes": 0, "replies": [] }, { "id": 1302960, "author_name": "MichaelP", "author_url": "", "post_date": "2021-05-11T20:01:20.170000", "content": "
\n

my team member @joven1997 has 16 x V100 GPUs in the DGX box. So i used it

\n
\n

Wow - that's pretty sweet @morizin - were you guys able to utilise all that GPU compute? How many models were you able to run concurrently and how did you keep track of all your results? Good luck with the last few hours.

", "votes": 0, "replies": [] }, { "id": 1302976, "author_name": "Mohammed Rizin V K", "author_url": "", "post_date": "2021-05-11T20:19:58.533000", "content": "

we didnt use all of them at a time. we used only 6 to 8 of them. We were training fold networks. but sad, this 16 x V100 GPU cannot be used in kaggle submission.

", "votes": 0, "replies": [] }, { "id": 1302991, "author_name": "wayfarer", "author_url": "", "post_date": "2021-05-11T20:35:55.757000", "content": "

@datahobbit Most of the models of our team were trained using Kaggle TPU (Keras+TF), models trained on DGX were Pytorch based, the number of models runned concurently usually were 2-3. We used https://wandb.ai/ to track results. So basically DGX was not involved much. But in general the number of models running in the same time depends on the image size, batch size and model size.

", "votes": 1, "replies": [] }, { "id": 1305647, "author_name": "MichaelP", "author_url": "", "post_date": "2021-05-13T12:08:25.263000", "content": "

@morizin and @joven1997 thank you for the explanation and congratulations on your strong result. Too bad it wasn't quite gold, I know how disappointing it can be to get so close and miss out…
\nIn a lot of competitions, I have been pretty maxed out on the hardware I had. I also know that even with much less hardware existing results and planning what to run next needed some careful thought. That's why I have wondered if I had a lot more hardware available, would I actually be able to use all of it? Your answers gave some interesting insights into that. Good luck in future competitions!

", "votes": 1, "replies": [] } ] }, { "id": 1303780, "author_name": "Salim Khazem", "author_url": "", "post_date": "2021-05-12T08:32:10.610000", "content": "

Kaggle / Colab, i m waiting for availability of RTX 3090 to build workstation

", "votes": 1, "replies": [] }, { "id": 1303500, "author_name": "Ivan Ilyushchenko", "author_url": "", "post_date": "2021-05-12T05:14:41.643000", "content": "

GeForce 970 GTX/24GB RAM/simple HDD for single cell classification model based on EfficientNetV2. It trained about 9 days with some simple optimization tricks.

", "votes": 1, "replies": [] }, { "id": 1301832, "author_name": "Raman", "author_url": "", "post_date": "2021-05-11T09:06:44.130000", "content": "

2x GTX 1080 Ti, i7-8700K, 32 GB RAM (with a 32 GB swap 😅)
\noccasionally I used RTX 2080 Ti on another similar gaming/DL PC

", "votes": 1, "replies": [] }, { "id": 1303519, "author_name": "Mikhail Gurevich", "author_url": "", "post_date": "2021-05-12T05:38:20.220000", "content": "

Home GTX1080/16GB RAM + 1TB SSD + extensive use of Colab Pro :-)

", "votes": 2, "replies": [] }, { "id": 1302758, "author_name": "sin", "author_url": "", "post_date": "2021-05-11T17:35:42.763000", "content": "

My sincere suggestion for you: set up your pc in a way that it reaches a good balance of brainstorming speed and experiment speed. It’s also important to take how much time you can devote to competitions. In fact, I think your setup should be able to play in most of the competitions.

", "votes": 2, "replies": [] }, { "id": 1301865, "author_name": "cool_rabbit", "author_url": "", "post_date": "2021-05-11T09:26:11.740000", "content": "

Kaggle Notebook and Datasets (TPU training & GPU inference)

", "votes": 2, "replies": [] } ], "raw_markdown_by_id": { "1301719": "What are you guys using? What's a standard/prefered setup for grandmasters?\n\nI have a gaming PC with a RTX 3090, i7-10700KF and 64GB RAM. Wasn't easy to find just the GPU alone 2-3 motnhs ago so I just bought it together. I did buy 3 extra SSD-s after for this contest ... never enough space :D\n\nI would love to hear suggestions/preferences on setups for computer vision contests...", "1301776": "I used Kaggle TPU", "1301728": "Colab Pro, 10 USD/Month for V100 GPU :)", "1301828": "HP laptop, Kaggle GPU, Colab Pro", "1301748": "Colab Pro (TPU ) and Kaggle (TPU) ", "1302726": "Home RTX 3090/64GB RAM, 1TB SSD.", "1301871": "my team member @joven1997 has 16 x V100 GPUs in the DGX box. So i used it", "1303780": "Kaggle / Colab, i m waiting for availability of RTX 3090 to build workstation ", "1303500": "GeForce 970 GTX/24GB RAM/simple HDD for single cell classification model based on EfficientNetV2. It trained about 9 days with some simple optimization tricks.", "1301832": "2x GTX 1080 Ti, i7-8700K, 32 GB RAM (with a 32 GB swap 😅)\noccasionally I used RTX 2080 Ti on another similar gaming/DL PC", "1303519": "Home GTX1080/16GB RAM + 1TB SSD + extensive use of Colab Pro :-)", "1302758": "My sincere suggestion for you: set up your pc in a way that it reaches a good balance of brainstorming speed and experiment speed. It’s also important to take how much time you can devote to competitions. In fact, I think your setup should be able to play in most of the competitions.", "1301865": "Kaggle Notebook and Datasets (TPU training & GPU inference)" } }