{ "id": 190643, "title": "GPU vs CPU time the same?", "url": "/competitions/rsna-str-pulmonary-embolism-detection/discussion/190643", "author_name": "kaaaaaaaagglleerrr", "post_date": "2020-10-12T19:27:26.944000", "votes": -1, "comment_count": 5, "views": 0, "content": "

Usually these are not the same number. Is this a typo in the code-requirements point?
\n https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection/overview/code-requirements

\n
\n

CPU Notebook <= 9 hours run-time
\n GPU Notebook <= 9 hours run-time

\n
", "messages": [ { "id": 1047636, "postDate": "2020-10-12T19:27:26.943Z", "content": "

Usually these are not the same number. Is this a typo in the code-requirements point?
\n https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection/overview/code-requirements

\n
\n

CPU Notebook <= 9 hours run-time
\n GPU Notebook <= 9 hours run-time

\n
", "rawMarkdown": "Usually these are not the same number. Is this a typo in the code-requirements point?\n https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection/overview/code-requirements\n> CPU Notebook <= 9 hours run-time\n> GPU Notebook <= 9 hours run-time", "votes": -1 }, { "id": 1047656, "postDate": "2020-10-12T20:12:52.020Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 1047674, "postDate": "2020-10-12T20:47:15.403Z", "content": "

I am not certain you can run longer than 9 hours on the private test. My Notebook gets stopped at approximately the nine hour mark with a time-out error.

", "rawMarkdown": "I am not certain you can run longer than 9 hours on the private test. My Notebook gets stopped at approximately the nine hour mark with a time-out error.", "votes": 2 }, { "id": 1047676, "postDate": "2020-10-12T20:49:15.220Z", "content": "

Hmm yep you're right, I seem to be running into issues around 9 hours, \"Notebook Timeout\". My mistake, not sure why I thought that.

", "rawMarkdown": "Hmm yep you're right, I seem to be running into issues around 9 hours, \"Notebook Timeout\". My mistake, not sure why I thought that. " }, { "id": 1047697, "postDate": "2020-10-12T21:21:15.643Z", "content": "

Me too, I confirm. I noticed that when turning on GPU, there is a message with a warning mentioning that GPU will shorten time processing for image processing and neural networks only, which means, to my understanding, that it will not improve time processing for other algorithms. In what I am concerned, I did not obtain my best result with NN.

", "rawMarkdown": "Me too, I confirm. I noticed that when turning on GPU, there is a message with a warning mentioning that GPU will shorten time processing for image processing and neural networks only, which means, to my understanding, that it will not improve time processing for other algorithms. In what I am concerned, I did not obtain my best result with NN.", "replies": [ { "id": 1047788, "postDate": "2020-10-13T00:10:09.317Z", "content": "

Typically a TPU or GPU will help CNN processing (Convolutional Neural Networks). So pytorch, keras, tensorflow - image processing models will get significant speed improvement (ResNet, DenseNet, EfficientNet, etc).

\n

For other types of models, you'll have to look into each model and see if TPU or GPU will help.

", "rawMarkdown": "Typically a TPU or GPU will help CNN processing (Convolutional Neural Networks). So pytorch, keras, tensorflow - image processing models will get significant speed improvement (ResNet, DenseNet, EfficientNet, etc).\n\nFor other types of models, you'll have to look into each model and see if TPU or GPU will help." } ] } ] } ], "comments": [ { "id": 1047656, "author_name": "", "author_url": "", "post_date": "2020-10-12T20:12:52.020000", "content": "", "votes": 0, "replies": [ { "id": 1047674, "author_name": "quadcore/Richard Epstein", "author_url": "", "post_date": "2020-10-12T20:47:15.403000", "content": "

I am not certain you can run longer than 9 hours on the private test. My Notebook gets stopped at approximately the nine hour mark with a time-out error.

", "votes": 2, "replies": [] }, { "id": 1047676, "author_name": "Stanley Zheng", "author_url": "", "post_date": "2020-10-12T20:49:15.220000", "content": "

Hmm yep you're right, I seem to be running into issues around 9 hours, \"Notebook Timeout\". My mistake, not sure why I thought that.

", "votes": 0, "replies": [] }, { "id": 1047697, "author_name": "Catadanna", "author_url": "", "post_date": "2020-10-12T21:21:15.643000", "content": "

Me too, I confirm. I noticed that when turning on GPU, there is a message with a warning mentioning that GPU will shorten time processing for image processing and neural networks only, which means, to my understanding, that it will not improve time processing for other algorithms. In what I am concerned, I did not obtain my best result with NN.

", "votes": 0, "replies": [ { "id": 1047788, "author_name": "quadcore/Richard Epstein", "author_url": "", "post_date": "2020-10-13T00:10:09.317000", "content": "

Typically a TPU or GPU will help CNN processing (Convolutional Neural Networks). So pytorch, keras, tensorflow - image processing models will get significant speed improvement (ResNet, DenseNet, EfficientNet, etc).

\n

For other types of models, you'll have to look into each model and see if TPU or GPU will help.

", "votes": 0, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "1047636": "Usually these are not the same number. Is this a typo in the code-requirements point?\n https://www.kaggle.com/c/rsna-str-pulmonary-embolism-detection/overview/code-requirements\n> CPU Notebook <= 9 hours run-time\n> GPU Notebook <= 9 hours run-time", "1047656": "" } }