{ "id": 192341, "title": "Anyone using Google Cloud?", "url": "/competitions/rsna-str-pulmonary-embolism-detection/discussion/192341", "author_name": "Dewei Chen", "post_date": "2020-10-21T04:06:26.520000", "votes": 4, "comment_count": 14, "views": 0, "content": "

Our GPU has 30 hours limit per week.

\n

I am almost run out of my weekly quote, and a Google Cloud block said if using Google Cloud notebook, there is no more time limit, and I notice that Cloud has 300USD free try. Does anyone try this?
\nIf I wish to keep using the Kaggle environment notebook, how much should I pay monthly? May I run two GPU notebook after buying this server?

\n

Thanks very much for your reply.

", "messages": [ { "id": 1055734, "postDate": "2020-10-21T04:06:26.520Z", "content": "

Our GPU has 30 hours limit per week.

\n

I am almost run out of my weekly quote, and a Google Cloud block said if using Google Cloud notebook, there is no more time limit, and I notice that Cloud has 300USD free try. Does anyone try this?
\nIf I wish to keep using the Kaggle environment notebook, how much should I pay monthly? May I run two GPU notebook after buying this server?

\n

Thanks very much for your reply.

", "rawMarkdown": "Our GPU has 30 hours limit per week.\n\nI am almost run out of my weekly quote, and a Google Cloud block said if using Google Cloud notebook, there is no more time limit, and I notice that Cloud has 300USD free try. Does anyone try this? \nIf I wish to keep using the Kaggle environment notebook, how much should I pay monthly? May I run two GPU notebook after buying this server? \n\nThanks very much for your reply.", "votes": 4 }, { "id": 1055923, "postDate": "2020-10-21T08:28:28.487Z", "content": "

Hi @dwchen take a look at this discussion thread which should answer one of your questions.

\n

With regards to running two GPU notebooks, you can use a GPU and TPU notebook at the same time on Kaggle.

\n

Hope this helps!

", "rawMarkdown": "Hi @dwchen take a look at this [discussion thread](https://www.kaggle.com/c/microsoft-malware-prediction/discussion/81034) which should answer one of your questions.\n\nWith regards to running two GPU notebooks, you can use a GPU and TPU notebook at the same time on Kaggle.\n\nHope this helps!", "votes": 1, "replies": [ { "id": 1055937, "postDate": "2020-10-21T08:52:50.963Z", "content": "

Thanks for the reply, but we cannot run GPU training code directly in the TPU environment right? TPU needs specific instructions to activate it.

", "rawMarkdown": "Thanks for the reply, but we cannot run GPU training code directly in the TPU environment right? TPU needs specific instructions to activate it." }, { "id": 1055939, "postDate": "2020-10-21T09:02:05.550Z", "content": "

Yes you have to add code to use the TPU, can view my notebook with it implemented to switch on and off the TPU code.

\n

However although it works the TPU performance would be approximately equivalent to the GPU due to me keeping the batch size the same. I never got around to figuring out how to optimise the model learn rate and batch size to maximise performance, but it does what I wanted it to do at the time which is to essentially have the ability to have the equivalent of two GPU notebooks.

", "rawMarkdown": "Yes you have to add code to use the TPU, can view my [notebook](https://www.kaggle.com/manchunhui/openvaccine-gru-lstm-with-cnn) with it implemented to switch on and off the TPU code. \n\nHowever although it works the TPU performance would be approximately equivalent to the GPU due to me keeping the batch size the same. I never got around to figuring out how to optimise the model learn rate and batch size to maximise performance, but it does what I wanted it to do at the time which is to essentially have the ability to have the equivalent of two GPU notebooks.", "votes": 1 }, { "id": 1055946, "postDate": "2020-10-21T09:11:26.103Z", "content": "

Thanks, I'll learn your notebook.

", "rawMarkdown": "Thanks, I'll learn your notebook.", "votes": 1 } ] }, { "id": 1056800, "postDate": "2020-10-22T04:53:55.107Z", "rawMarkdown": "", "votes": 1, "isDeleted": true, "replies": [ { "id": 1056805, "postDate": "2020-10-22T05:02:11.730Z", "content": "

Did you make a copy of the notebook and instance it on Google Cloud?

", "rawMarkdown": "Did you make a copy of the notebook and instance it on Google Cloud? " }, { "id": 1056818, "postDate": "2020-10-22T05:22:55.293Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 1056839, "postDate": "2020-10-22T06:10:17.720Z", "content": "

The main reason is our GPU ran out of limit, that button means copy your code and environment to Google Cloud. After we create an instance at Google Cloud, we still need to pay for extra device running fee: A standard CPU and a Tesla V100 16G RAM.

", "rawMarkdown": "The main reason is our GPU ran out of limit, that button means copy your code and environment to Google Cloud. After we create an instance at Google Cloud, we still need to pay for extra device running fee: A standard CPU and a Tesla V100 16G RAM. " }, { "id": 1057853, "postDate": "2020-10-23T04:55:10.700Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 1057872, "postDate": "2020-10-23T05:17:15.933Z", "content": "

You can read the top comment, you can take a look at his notebook. I success to implement TPU in my notebook for training a desnet and also my local PC for training a efficientnet.

\n

Google Cloud tells me it will cost 1k USD for one month with a Tesla 16G RAM device. This quite expensive for personal use.

", "rawMarkdown": "You can read the top comment, you can take a look at his [notebook](https://www.kaggle.com/manchunhui/openvaccine-gru-lstm-with-cnn). I success to implement TPU in my notebook for training a desnet and also my local PC for training a efficientnet.\n\nGoogle Cloud tells me it will cost 1k USD for one month with a Tesla 16G RAM device. This quite expensive for personal use.\n" }, { "id": 1058228, "postDate": "2020-10-23T13:29:37.667Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 1058295, "postDate": "2020-10-23T14:40:16.690Z", "content": "

efficientnet, at the moment

", "rawMarkdown": "efficientnet, at the moment" }, { "id": 1059066, "postDate": "2020-10-24T15:15:17.647Z", "rawMarkdown": "", "isDeleted": true }, { "id": 1059369, "postDate": "2020-10-25T02:32:57.850Z", "content": "

About 1 to 3.

", "rawMarkdown": "About 1 to 3." } ] } ], "comments": [ { "id": 1055923, "author_name": "Manch Hui", "author_url": "", "post_date": "2020-10-21T08:28:28.487000", "content": "

Hi @dwchen take a look at this discussion thread which should answer one of your questions.

\n

With regards to running two GPU notebooks, you can use a GPU and TPU notebook at the same time on Kaggle.

\n

Hope this helps!

", "votes": 1, "replies": [ { "id": 1055937, "author_name": "Dewei Chen", "author_url": "", "post_date": "2020-10-21T08:52:50.963000", "content": "

Thanks for the reply, but we cannot run GPU training code directly in the TPU environment right? TPU needs specific instructions to activate it.

", "votes": 0, "replies": [] }, { "id": 1055939, "author_name": "Manch Hui", "author_url": "", "post_date": "2020-10-21T09:02:05.550000", "content": "

Yes you have to add code to use the TPU, can view my notebook with it implemented to switch on and off the TPU code.

\n

However although it works the TPU performance would be approximately equivalent to the GPU due to me keeping the batch size the same. I never got around to figuring out how to optimise the model learn rate and batch size to maximise performance, but it does what I wanted it to do at the time which is to essentially have the ability to have the equivalent of two GPU notebooks.

", "votes": 1, "replies": [] }, { "id": 1055946, "author_name": "Dewei Chen", "author_url": "", "post_date": "2020-10-21T09:11:26.103000", "content": "

Thanks, I'll learn your notebook.

", "votes": 1, "replies": [] } ] }, { "id": 1056800, "author_name": "", "author_url": "", "post_date": "2020-10-22T04:53:55.107000", "content": "", "votes": 1, "replies": [ { "id": 1056805, "author_name": "Dewei Chen", "author_url": "", "post_date": "2020-10-22T05:02:11.730000", "content": "

Did you make a copy of the notebook and instance it on Google Cloud?

", "votes": 0, "replies": [] }, { "id": 1056818, "author_name": "", "author_url": "", "post_date": "2020-10-22T05:22:55.293000", "content": "", "votes": 1, "replies": [] }, { "id": 1056839, "author_name": "Dewei Chen", "author_url": "", "post_date": "2020-10-22T06:10:17.720000", "content": "

The main reason is our GPU ran out of limit, that button means copy your code and environment to Google Cloud. After we create an instance at Google Cloud, we still need to pay for extra device running fee: A standard CPU and a Tesla V100 16G RAM.

", "votes": 0, "replies": [] }, { "id": 1057853, "author_name": "", "author_url": "", "post_date": "2020-10-23T04:55:10.700000", "content": "", "votes": 1, "replies": [] }, { "id": 1057872, "author_name": "Dewei Chen", "author_url": "", "post_date": "2020-10-23T05:17:15.933000", "content": "

You can read the top comment, you can take a look at his notebook. I success to implement TPU in my notebook for training a desnet and also my local PC for training a efficientnet.

\n

Google Cloud tells me it will cost 1k USD for one month with a Tesla 16G RAM device. This quite expensive for personal use.

", "votes": 0, "replies": [] }, { "id": 1058228, "author_name": "", "author_url": "", "post_date": "2020-10-23T13:29:37.667000", "content": "", "votes": 1, "replies": [] }, { "id": 1058295, "author_name": "Dewei Chen", "author_url": "", "post_date": "2020-10-23T14:40:16.690000", "content": "

efficientnet, at the moment

", "votes": 0, "replies": [] }, { "id": 1059066, "author_name": "", "author_url": "", "post_date": "2020-10-24T15:15:17.647000", "content": "", "votes": 0, "replies": [] }, { "id": 1059369, "author_name": "Dewei Chen", "author_url": "", "post_date": "2020-10-25T02:32:57.850000", "content": "

About 1 to 3.

", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "1055734": "Our GPU has 30 hours limit per week.\n\nI am almost run out of my weekly quote, and a Google Cloud block said if using Google Cloud notebook, there is no more time limit, and I notice that Cloud has 300USD free try. Does anyone try this? \nIf I wish to keep using the Kaggle environment notebook, how much should I pay monthly? May I run two GPU notebook after buying this server? \n\nThanks very much for your reply.", "1055923": "Hi @dwchen take a look at this [discussion thread](https://www.kaggle.com/c/microsoft-malware-prediction/discussion/81034) which should answer one of your questions.\n\nWith regards to running two GPU notebooks, you can use a GPU and TPU notebook at the same time on Kaggle.\n\nHope this helps!", "1056800": "" } }