Hi, I am wondering what platforms are you using to work on this huge data sets, would appreciate if you can guide for this competition.
", "messages": [ { "id": "221319", "postDate": "09/14/2017 19:53:17", "content": "Hi, I am wondering what platforms are you using to work on this huge data sets, would appreciate if you can guide for this competition.
", "rawMarkdown": "Hi, I am wondering what platforms are you using to work on this huge data sets, would appreciate if you can guide for this competition.", "votes": null }, { "id": "221322", "postDate": "09/14/2017 20:05:00", "content": "Same question! Is it Cost-effective to use the Amazon platform?
", "rawMarkdown": "Same question! Is it Cost-effective to use the Amazon platform?", "votes": null }, { "id": "221331", "postDate": "09/14/2017 20:33:16", "content": "Both the train and test files can be read and processed iteratively from disk. So, RAM shouldn't be much of a limiter. I did, though, use 64 cores when creating the image similarity benchmark submission. Total time was less than 12 hours. I didn't do any neural net exploration, so I can't comment about that.
", "rawMarkdown": "Both the train and test files can be read and processed iteratively from disk. So, RAM shouldn't be much of a limiter. I did, though, use 64 cores when creating the image similarity benchmark submission. Total time was less than 12 hours. I didn't do any neural net exploration, so I can't comment about that.", "votes": null }, { "id": "221337", "postDate": "09/14/2017 20:57:09", "content": "Can you make torrent download available for the data of this competition? It will be easier for us to download the data, especially if we have a slow broadband connexion, like me :(
", "rawMarkdown": "Can you make torrent download available for the data of this competition? It will be easier for us to download the data, especially if we have a slow broadband connexion, like me :(", "votes": null }, { "id": "221339", "postDate": "09/14/2017 21:06:38", "content": "It's cheaper to buy a gpu. Try the GCP free trial if you want, but 1080/1080Ti should work well for just about any contest.
", "rawMarkdown": "It's cheaper to buy a gpu. Try the GCP free trial if you want, but 1080/1080Ti should work well for just about any contest.", "votes": null }, { "id": "221342", "postDate": "09/14/2017 21:15:50", "content": "@Steven: Do you think that only one GPU will be sufficient for this competition?
", "rawMarkdown": "Steven: Do you think that only one GPU will be sufficient for this competition?", "votes": null }, { "id": "221478", "postDate": "09/15/2017 12:17:16", "content": "local 1080Ti has been shown to outperform cloud-based research grade GPUs in some tasks [citation needed]
", "rawMarkdown": "local 1080Ti has been shown to outperform cloud-based research grade GPUs in some tasks [citation needed]", "votes": null }, { "id": "221875", "postDate": "09/16/2017 16:15:57", "content": "do i need to pay for server class of machine from aws to train this huge datasets?
", "rawMarkdown": "do i need to pay for server class of machine from aws to train this huge datasets?", "votes": null }, { "id": "222136", "postDate": "09/17/2017 17:56:57", "content": "You don't need citation, GTX 1800 Ti has 11.3 TFLOPS FP32 -single precision, while K80 8.7.
", "rawMarkdown": "You don't need citation, GTX 1800 Ti has [11.3 TFLOPS FP32 -single precision](http://www.anandtech.com/show/11172/nvidia-unveils-geforce-gtx-1080-ti-next-week-699), while [K80 8.7](http://images.nvidia.com/content/tesla/pdf/nvidia-tesla-k80-overview.pdf).", "votes": null } ], "comments": [ { "id": 221322, "author_name": "daikinban", "author_url": "", "post_date": "09/14/2017 20:05:00", "content": "Same question! Is it Cost-effective to use the Amazon platform?
", "votes": null, "replies": [ { "id": 221339, "author_name": "stevenknguyen", "author_url": "", "post_date": "09/14/2017 21:06:38", "content": "It's cheaper to buy a gpu. Try the GCP free trial if you want, but 1080/1080Ti should work well for just about any contest.
", "votes": null, "replies": [] }, { "id": 221342, "author_name": "thomasseleck", "author_url": "", "post_date": "09/14/2017 21:15:50", "content": "@Steven: Do you think that only one GPU will be sufficient for this competition?
", "votes": null, "replies": [] }, { "id": 221478, "author_name": "gw00207", "author_url": "", "post_date": "09/15/2017 12:17:16", "content": "local 1080Ti has been shown to outperform cloud-based research grade GPUs in some tasks [citation needed]
", "votes": null, "replies": [] }, { "id": 222136, "author_name": "adamszalucha", "author_url": "", "post_date": "09/17/2017 17:56:57", "content": "You don't need citation, GTX 1800 Ti has 11.3 TFLOPS FP32 -single precision, while K80 8.7.
", "votes": null, "replies": [] } ] }, { "id": 221331, "author_name": "inversion", "author_url": "", "post_date": "09/14/2017 20:33:16", "content": "Both the train and test files can be read and processed iteratively from disk. So, RAM shouldn't be much of a limiter. I did, though, use 64 cores when creating the image similarity benchmark submission. Total time was less than 12 hours. I didn't do any neural net exploration, so I can't comment about that.
", "votes": null, "replies": [] }, { "id": 221337, "author_name": "thomasseleck", "author_url": "", "post_date": "09/14/2017 20:57:09", "content": "Can you make torrent download available for the data of this competition? It will be easier for us to download the data, especially if we have a slow broadband connexion, like me :(
", "votes": null, "replies": [] }, { "id": 221875, "author_name": "ncelq1273", "author_url": "", "post_date": "09/16/2017 16:15:57", "content": "do i need to pay for server class of machine from aws to train this huge datasets?
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "221319": "Hi, I am wondering what platforms are you using to work on this huge data sets, would appreciate if you can guide for this competition.", "221322": "Same question! Is it Cost-effective to use the Amazon platform?", "221331": "Both the train and test files can be read and processed iteratively from disk. So, RAM shouldn't be much of a limiter. I did, though, use 64 cores when creating the image similarity benchmark submission. Total time was less than 12 hours. I didn't do any neural net exploration, so I can't comment about that.", "221337": "Can you make torrent download available for the data of this competition? It will be easier for us to download the data, especially if we have a slow broadband connexion, like me :(", "221339": "It's cheaper to buy a gpu. Try the GCP free trial if you want, but 1080/1080Ti should work well for just about any contest.", "221342": "Steven: Do you think that only one GPU will be sufficient for this competition?", "221478": "local 1080Ti has been shown to outperform cloud-based research grade GPUs in some tasks [citation needed]", "221875": "do i need to pay for server class of machine from aws to train this huge datasets?", "222136": "You don't need citation, GTX 1800 Ti has [11.3 TFLOPS FP32 -single precision](http://www.anandtech.com/show/11172/nvidia-unveils-geforce-gtx-1080-ti-next-week-699), while [K80 8.7](http://images.nvidia.com/content/tesla/pdf/nvidia-tesla-k80-overview.pdf)." }, "source": "meta" }