![\"\"](\"https://projectstoreph.com/wp-content/uploads/2019/03/IMG_20190313_234306.jpg\")
I am wondering a lot of folks would be using their own infrastructures for this competition. How does the hardware specs look like?
\n\nRAM? SSD? GPU? Cloud instance? TPU? ...........\n
I am wondering a lot of folks would be using their own infrastructures for this competition. How does the hardware specs look like?
\n\nRAM? SSD? GPU? Cloud instance? TPU? ...........\n
Xenon box with 64GB +64GB swap, Ubuntu Linux 18.04, 2xNVidia RTX2080ti, 500GB SSD + 4TB HDD, hybrid liquid cooling, Intel i9 with 10 cores/20 threads. For messing around Lenovo T460, 12GB Ram, no GPU, Ubuntu Linux.
", "rawMarkdown": "Xenon box with 64GB +64GB swap, Ubuntu Linux 18.04, 2xNVidia RTX2080ti, 500GB SSD + 4TB HDD, hybrid liquid cooling, Intel i9 with 10 cores/20 threads. For messing around Lenovo T460, 12GB Ram, no GPU, Ubuntu Linux.", "votes": null }, { "id": "550138", "postDate": "06/11/2019 10:40:13", "content": "I have a home PC with i7-8700K, 64 Gb RAM and 1 1080ti.
", "rawMarkdown": "I have a home PC with i7-8700K, 64 Gb RAM and 1 1080ti.", "votes": null }, { "id": "550140", "postDate": "06/11/2019 10:44:12", "content": "Threadripper 2950X, RTX-2080ti, GTX-1070ti, 128Gb RAM
", "rawMarkdown": "Threadripper 2950X, RTX-2080ti, GTX-1070ti, 128Gb RAM", "votes": null }, { "id": "550142", "postDate": "06/11/2019 10:48:07", "content": "👍 Wow thats a serious machine. I guess 15 grand plus.
", "rawMarkdown": "👍 Wow thats a serious machine. I guess 15 grand plus.", "votes": null }, { "id": "550171", "postDate": "06/11/2019 11:27:43", "content": "Nah, half that and was pre-configured for me as well. I used to use a cloud instance with GPU on Azure but the cost of that in the long term was just not worth it. Having a decent home-office setup will last me a long time and it will pay for itself (compared to cloud). My company does blockchain anyway so plenty of other ways it is useful as well, and if I get half-good at ML and start consulting I'll need the tools of the trade.
", "rawMarkdown": "Nah, half that and was pre-configured for me as well. I used to use a cloud instance with GPU on Azure but the cost of that in the long term was just not worth it. Having a decent home-office setup will last me a long time and it will pay for itself (compared to cloud). My company does blockchain anyway so plenty of other ways it is useful as well, and if I get half-good at ML and start consulting I'll need the tools of the trade.", "votes": null }, { "id": "550172", "postDate": "06/11/2019 11:28:37", "content": "Awesome setup!
", "rawMarkdown": "Awesome setup!", "votes": null }, { "id": "550185", "postDate": "06/11/2019 11:44:18", "content": "Have same as Andrew but with two 1080. Have tried to get lightgbm running on GPU but with no success.
", "rawMarkdown": "Have same as Andrew but with two 1080. Have tried to get lightgbm running on GPU but with no success.", "votes": null }, { "id": "550231", "postDate": "06/11/2019 12:22:38", "content": "currently laptop i5 8gb ram nvidia 940mx :)\nwhen I hit the memory limit then I will move to the cloud
", "rawMarkdown": "currently laptop i5 8gb ram nvidia 940mx :)\nwhen I hit the memory limit then I will move to the cloud", "votes": null }, { "id": "550271", "postDate": "06/11/2019 13:04:03", "content": "Are you on windows? I have a similar setup and absolutely could not get lightgbm to build with GPU support (or at all, really).
\n\nHowever, on my Linux machine building it for gpu worked like a charm.
", "rawMarkdown": "Are you on windows? I have a similar setup and absolutely could not get lightgbm to build with GPU support (or at all, really). \n\nHowever, on my Linux machine building it for gpu worked like a charm.", "votes": null }, { "id": "550514", "postDate": "06/11/2019 17:32:56", "content": "Intel 4790K, 16 Gig RAM and a 1070Ti.
", "rawMarkdown": "Intel 4790K, 16 Gig RAM and a 1070Ti.", "votes": null }, { "id": "550549", "postDate": "06/11/2019 18:23:14", "content": "laptop Intel Core i7-8550U 16 gb ram
", "rawMarkdown": "laptop Intel Core i7-8550U 16 gb ram", "votes": null }, { "id": "550598", "postDate": "06/11/2019 19:46:21", "content": "Intel 3570K, 16GB RAM, RTX2070, running Ubuntu. I also still have a bunch of free credit for Google Cloud so I used this to generate some graph features. I've given up on competing though.
", "rawMarkdown": "Intel 3570K, 16GB RAM, RTX2070, running Ubuntu. I also still have a bunch of free credit for Google Cloud so I used this to generate some graph features. I've given up on competing though.", "votes": null }, { "id": "550683", "postDate": "06/11/2019 23:20:03", "content": "9600k, 32 GB RAM, RTX2060
", "rawMarkdown": "9600k, 32 GB RAM, RTX2060", "votes": null }, { "id": "550804", "postDate": "06/12/2019 03:32:03", "content": "https://www.youtube.com/watch?v=_QgexYD-tZg
\n\nI7 6700K,\n32 GB DDR4\n1080 TI NVIDIA\nM.2 SSD
\n\nRUNNING: MAC OSX HIGH SIERRA (YES, IM A HACKINTOSHER XD)
", "rawMarkdown": "https://www.youtube.com/watch?v=_QgexYD-tZg\n\nI7 6700K,\n32 GB DDR4\n1080 TI NVIDIA\nM.2 SSD \n\nRUNNING: MAC OSX HIGH SIERRA (YES, IM A HACKINTOSHER XD)", "votes": null }, { "id": "550939", "postDate": "06/12/2019 06:43:14", "content": "i tried setting up a environment on floydhub, but i am not able to get the gpu support going \nLightGBMError: GPU Tree Learner was not enabled in this build.\nPlease recompile with CMake option -DUSE_GPU=1
\n\nin the long run, having your own rig pays off
", "rawMarkdown": "i tried setting up a environment on floydhub, but i am not able to get the gpu support going \nLightGBMError: GPU Tree Learner was not enabled in this build.\nPlease recompile with CMake option -DUSE_GPU=1\n\nin the long run, having your own rig pays off", "votes": null }, { "id": "551112", "postDate": "06/12/2019 11:02:03", "content": "Intel Xeon E5-2637 @ 3.50GHz (16 cores)\n128 GB RAM\nNVidia Quadro M4000
", "rawMarkdown": "Intel Xeon E5-2637 @ 3.50GHz (16 cores)\n128 GB RAM\nNVidia Quadro M4000", "votes": null }, { "id": "551184", "postDate": "06/12/2019 12:33:34", "content": "I gave up on my Hackintosh... I just remote in from my macbook :)
", "rawMarkdown": "I gave up on my Hackintosh... I just remote in from my macbook :)", "votes": null }, { "id": "551185", "postDate": "06/12/2019 12:33:39", "content": "Ubuntu 18.04\nIntel(R) Core(TM) i9-9820X CPU Overclocked to 4.1 GHz\n64 GB RAM\nRTX 2080ti
", "rawMarkdown": "Ubuntu 18.04\nIntel(R) Core(TM) i9-9820X CPU Overclocked to 4.1 GHz\n64 GB RAM\nRTX 2080ti", "votes": null }, { "id": "551233", "postDate": "06/12/2019 13:44:50", "content": "Ubuntu 18.04\nThread Ripper\n128GB RAM\nTitan Xp x2\nNVME 1Gb
", "rawMarkdown": "Ubuntu 18.04\nThread Ripper\n128GB RAM\nTitan Xp x2\nNVME 1Gb", "votes": null }, { "id": "551405", "postDate": "06/12/2019 17:17:57", "content": "Well, after reading through all this I have a feeling that my hardware is from the stone age.
", "rawMarkdown": "Well, after reading through all this I have a feeling that my hardware is from the stone age.", "votes": null }, { "id": "552617", "postDate": "06/14/2019 07:59:27", "content": "", "rawMarkdown": "", "votes": null }, { "id": "552787", "postDate": "06/14/2019 14:18:16", "content": "2 machines each \n28 Xenon Cores with 56 threads\n256 GB Ram
\n\n1 DGX-1\n512 GB RAM\n8 Tesla V 100 32Gb each\nDual 20-Core Intel Xeon E5-2698 v4 2.2 GHz
", "rawMarkdown": "2 machines each \n28 Xenon Cores with 56 threads\n256 GB Ram\n\n1 DGX-1\n512 GB RAM\n8 Tesla V 100 32Gb each\nDual 20-Core Intel Xeon E5-2698 v4 2.2 GHz", "votes": null }, { "id": "552809", "postDate": "06/14/2019 15:00:15", "content": "Wow, our company is so rich!!!!
", "rawMarkdown": "Wow, our company is so rich!!!!", "votes": null }, { "id": "552824", "postDate": "06/14/2019 15:53:58", "content": "Dell xps13\nManjaro 18\n4 cores Intel i5-7200U \n8GB RAM\n128G SSD\nNo GPU
", "rawMarkdown": "Dell xps13\nManjaro 18\n4 cores Intel i5-7200U \n8GB RAM\n128G SSD\nNo GPU", "votes": null }, { "id": "553125", "postDate": "06/15/2019 06:30:12", "content": "@abhinand05 lol
\n\nBut going through older Kaggle competitions people used dual cores or less. Its not always about brute-forcing and massive compute power, rather some clever and efficient coding. It is simply insane how people can come up with 10 liners, when others have to program an elaborate 1000 lines of code.
\n\nSee the solutions to the Google code challenge archives, its crazy:\nhttps://codingcompetitions.withgoogle.com/codejam/archive
\n\nYes, its nice to have a 100k GPU cluster and 1000 CPU cores, but nothing beats ingenuity. Kaggle kernel only challenges could equalize the field, basically if thousands of Kagglers sit in a single room. But then we have college entry exam situations. Not sure thats a lot of fun.
", "rawMarkdown": "abhinand05 lol\n\nBut going through older Kaggle competitions people used dual cores or less. Its not always about brute-forcing and massive compute power, rather some clever and efficient coding. It is simply insane how people can come up with 10 liners, when others have to program an elaborate 1000 lines of code.\n\nSee the solutions to the Google code challenge archives, its crazy:\nhttps://codingcompetitions.withgoogle.com/codejam/archive\n\nYes, its nice to have a 100k GPU cluster and 1000 CPU cores, but nothing beats ingenuity. Kaggle kernel only challenges could equalize the field, basically if thousands of Kagglers sit in a single room. But then we have college entry exam situations. Not sure thats a lot of fun.", "votes": null }, { "id": "553144", "postDate": "06/15/2019 07:23:31", "content": "Forcing Kagglers to use Python is not right. Get C# into kernels? (Oh, on topic: CPU 10 GHz (4 x 2.5), GPU: 288 GHz (480 x .6) and memory: 8 GB)
", "rawMarkdown": "Forcing Kagglers to use Python is not right. Get C# into kernels? (Oh, on topic: CPU 10 GHz (4 x 2.5), GPU: 288 GHz (480 x .6) and memory: 8 GB)", "votes": null }, { "id": "553153", "postDate": "06/15/2019 08:00:18", "content": "Wow
", "rawMarkdown": "Wow", "votes": null }, { "id": "553159", "postDate": "06/15/2019 08:13:37", "content": "Don't forget the metric suggests to can work on each type individually so this will reduce the problem space by 8
", "rawMarkdown": "Don't forget the metric suggests to can work on each type individually so this will reduce the problem space by 8", "votes": null }, { "id": "553176", "postDate": "06/15/2019 09:11:06", "content": "Not exactly, considering that different types have different amount of samples.
", "rawMarkdown": "Not exactly, considering that different types have different amount of samples.", "votes": null }, { "id": "553200", "postDate": "06/15/2019 09:53:15", "content": "True but you can do what you did - calculate fc as a whole then feed that into individual models
", "rawMarkdown": "True but you can do what you did - calculate fc as a whole then feed that into individual models", "votes": null }, { "id": "554265", "postDate": "06/17/2019 07:52:37", "content": "Standard iMac and MacbookAir
", "rawMarkdown": "Standard iMac and MacbookAir", "votes": null }, { "id": "554575", "postDate": "06/17/2019 17:18:13", "content": "I am getting some serious anxiety level lol. I bet anyone here to beat my specs:\nCompaq 610\nUbuntu 18.04\n2 GB RAM\nIntel Core 2 Duo
", "rawMarkdown": "I am getting some serious anxiety level lol. I bet anyone here to beat my specs:\nCompaq 610\nUbuntu 18.04\n2 GB RAM\nIntel Core 2 Duo", "votes": null }, { "id": "554585", "postDate": "06/17/2019 17:34:27", "content": "You should try using your phone ;)
", "rawMarkdown": "You should try using your phone ;)", "votes": null }, { "id": "554751", "postDate": "06/18/2019 01:41:55", "content": "just wow... what a beast...
", "rawMarkdown": "just wow... what a beast...", "votes": null }, { "id": "554794", "postDate": "06/18/2019 03:04:39", "content": "Haha, this has changed my perspective on the kind of specs other Kagglers are using, I thought it would be closer to Alienware, Predator or MSI. I use a Macbook Pro.
", "rawMarkdown": "Haha, this has changed my perspective on the kind of specs other Kagglers are using, I thought it would be closer to Alienware, Predator or MSI. I use a Macbook Pro.", "votes": null }, { "id": "554799", "postDate": "06/18/2019 03:11:44", "content": "yeah stick with ubuntu :)
", "rawMarkdown": "yeah stick with ubuntu :)", "votes": null }, { "id": "554867", "postDate": "06/18/2019 05:56:55", "content": "You should try using abacus 😆
", "rawMarkdown": "You should try using abacus 😆", "votes": null }, { "id": "555245", "postDate": "06/18/2019 16:15:36", "content": "Mine is too; 7 years old. I am currently not doing great in this contest but I did well in the earthquake one. Just goes to show you can compete without fancy hardware.
", "rawMarkdown": "Mine is too; 7 years old. I am currently not doing great in this contest but I did well in the earthquake one. Just goes to show you can compete without fancy hardware.", "votes": null }, { "id": "556797", "postDate": "06/20/2019 17:12:40", "content": "Haha, I am also using AMD Thread Ripper, I am using 1950X.
", "rawMarkdown": "Haha, I am also using AMD Thread Ripper, I am using 1950X.", "votes": null }, { "id": "556806", "postDate": "06/20/2019 17:23:30", "content": "I think we tie ;) Chromeos, 4 GB RAM, intel celeron. but i always use kernels
", "rawMarkdown": "I think we tie ;) Chromeos, 4 GB RAM, intel celeron. but i always use kernels", "votes": null }, { "id": "558732", "postDate": "06/22/2019 20:24:16", "content": "I'm gonna need a bigger boat...
", "rawMarkdown": "I'm gonna need a bigger boat...", "votes": null }, { "id": "566686", "postDate": "07/02/2019 13:09:44", "content": "Hi, @scirpus , @kingychiu , and @taniaj !
\n\nCould you share some of the specifications of your RAM and motherboard? 😆
\n\nHope this rather technical question does not bother you guys. I am building a computer also with Threadripper (1950x) now and plan to get 128 GB RAM. But I am a little confused by the CPU's memory specifications. Most forum discussions are for smaller RAM.
\n\nAccording to AMD website (https://www.amd.com/en/products/cpu/amd-ryzen-threadripper-2950x)\nSystem Memory Specification : 2933MHz\nSystem Memory Type : DDR4\nMemory Channels : 4
\n\nBut the more RAM slots we use, it seems the lower frequency it can achieve. According to \nhttps://en.wikichip.org/wiki/amd/ryzen_threadripper/2950x
\n\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1557202%2F69a3013d5f5798f35329bb5fe72919a3%2Fram.png?generation=1562071985388831&alt=media\")
I guess typically 128GB = 16GB * 8 and then it will be one of those \"8 of 8\" in the table above. Then, the operating frequency 2133MHz/1866MHz is much lower than the official website 2933MHz.
\n\nI am not sure what type of memory operating frequency to buy now. So, I think perhaps I can follow what you guys are using :)
\n\nThanks a lot!
", "rawMarkdown": "Hi, @scirpus , @kingychiu , and @taniaj !\n\nCould you share some of the specifications of your RAM and motherboard? 😆 \n\nHope this rather technical question does not bother you guys. I am building a computer also with Threadripper (1950x) now and plan to get 128 GB RAM. But I am a little confused by the CPU's memory specifications. Most forum discussions are for smaller RAM.\n\nAccording to AMD website (https://www.amd.com/en/products/cpu/amd-ryzen-threadripper-2950x)\nSystem Memory Specification : 2933MHz\nSystem Memory Type : DDR4\nMemory Channels : 4\n\nBut the more RAM slots we use, it seems the lower frequency it can achieve. According to \nhttps://en.wikichip.org/wiki/amd/ryzen_threadripper/2950x\n\n\n\nI guess typically 128GB = 16GB * 8 and then it will be one of those \"8 of 8\" in the table above. Then, the operating frequency 2133MHz/1866MHz is much lower than the official website 2933MHz.\n\nI am not sure what type of memory operating frequency to buy now. So, I think perhaps I can follow what you guys are using :)\n\nThanks a lot!", "votes": null }, { "id": "566695", "postDate": "07/02/2019 13:30:36", "content": "You will need 8 Ram slots. Here is how I managed to get 128GB Ram with 1950X. I built it during black friday 2018 on new egg :) (Sadly there weren't many discounts on GPUs)
\n\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F690886%2F1d0142cf7e85365c8088464787862fe9%2FCapture.PNG?generation=1562074038307170&alt=media\")
Board: GIGABYTE X399 AORUS XTREME\nRAM: Corsair DIMM 32GB DDR4-3000 Kit (x4)
", "rawMarkdown": "Board: GIGABYTE X399 AORUS XTREME\nRAM: Corsair DIMM 32GB DDR4-3000 Kit (x4)", "votes": null }, { "id": "566983", "postDate": "07/02/2019 21:59:20", "content": "Thanks, @kingychiu @taniaj !
\n\nI feel much more comfortable to make the purchase now!
\n\nBack Friday is good! 4th July is coming, but did not see much discount yet 🙌
", "rawMarkdown": "Thanks, @kingychiu @taniaj !\n\nI feel much more comfortable to make the purchase now!\n\nBack Friday is good! 4th July is coming, but did not see much discount yet 🙌", "votes": null }, { "id": "567028", "postDate": "07/03/2019 01:06:03", "content": "Make sure you spend money also on cooling if you will keep 100% CPU usage or even overclocking.
", "rawMarkdown": "Make sure you spend money also on cooling if you will keep 100% CPU usage or even overclocking.", "votes": null }, { "id": "571231", "postDate": "07/09/2019 10:49:05", "content": "\n\n\nI use a Macbook Pro.
\n
I fried a macbook pro with Kaggle. Sure, it was a mid 2015 and it is eligible in Apple recent recall. Batteries just blew out of proportion because of heat.
", "rawMarkdown": "> I use a Macbook Pro.\n\nI fried a macbook pro with Kaggle. Sure, it was a mid 2015 and it is eligible in Apple recent recall. Batteries just blew out of proportion because of heat.", "votes": null }, { "id": "571906", "postDate": "07/10/2019 08:21:11", "content": "I have the same question with you.I use Window10.But XGBoost and Tensorflow is workable running on GPU .
", "rawMarkdown": "I have the same question with you.I use Window10.But XGBoost and Tensorflow is workable running on GPU .", "votes": null }, { "id": "573170", "postDate": "07/11/2019 22:03:10", "content": "@kingychiu : what was the cost for each components on black friday sales??
", "rawMarkdown": "kingychiu : what was the cost for each components on black friday sales??", "votes": null } ], "comments": [ { "id": 550135, "author_name": "gideonvos", "author_url": "", "post_date": "06/11/2019 10:35:49", "content": "Xenon box with 64GB +64GB swap, Ubuntu Linux 18.04, 2xNVidia RTX2080ti, 500GB SSD + 4TB HDD, hybrid liquid cooling, Intel i9 with 10 cores/20 threads. For messing around Lenovo T460, 12GB Ram, no GPU, Ubuntu Linux.
", "votes": null, "replies": [ { "id": 550142, "author_name": "cyberia", "author_url": "", "post_date": "06/11/2019 10:48:07", "content": "👍 Wow thats a serious machine. I guess 15 grand plus.
", "votes": null, "replies": [] }, { "id": 550171, "author_name": "gideonvos", "author_url": "", "post_date": "06/11/2019 11:27:43", "content": "Nah, half that and was pre-configured for me as well. I used to use a cloud instance with GPU on Azure but the cost of that in the long term was just not worth it. Having a decent home-office setup will last me a long time and it will pay for itself (compared to cloud). My company does blockchain anyway so plenty of other ways it is useful as well, and if I get half-good at ML and start consulting I'll need the tools of the trade.
", "votes": null, "replies": [] } ] }, { "id": 550138, "author_name": "artgor", "author_url": "", "post_date": "06/11/2019 10:40:13", "content": "I have a home PC with i7-8700K, 64 Gb RAM and 1 1080ti.
", "votes": null, "replies": [ { "id": 550185, "author_name": "pcjimmmy", "author_url": "", "post_date": "06/11/2019 11:44:18", "content": "Have same as Andrew but with two 1080. Have tried to get lightgbm running on GPU but with no success.
", "votes": null, "replies": [] }, { "id": 550271, "author_name": "interneuron", "author_url": "", "post_date": "06/11/2019 13:04:03", "content": "Are you on windows? I have a similar setup and absolutely could not get lightgbm to build with GPU support (or at all, really).
\n\nHowever, on my Linux machine building it for gpu worked like a charm.
", "votes": null, "replies": [] }, { "id": 571906, "author_name": "fangwenjun", "author_url": "", "post_date": "07/10/2019 08:21:11", "content": "I have the same question with you.I use Window10.But XGBoost and Tensorflow is workable running on GPU .
", "votes": null, "replies": [] } ] }, { "id": 550140, "author_name": "taniaj", "author_url": "", "post_date": "06/11/2019 10:44:12", "content": "Threadripper 2950X, RTX-2080ti, GTX-1070ti, 128Gb RAM
", "votes": null, "replies": [ { "id": 550172, "author_name": "gideonvos", "author_url": "", "post_date": "06/11/2019 11:28:37", "content": "Awesome setup!
", "votes": null, "replies": [] } ] }, { "id": 550231, "author_name": "snovik1975", "author_url": "", "post_date": "06/11/2019 12:22:38", "content": "currently laptop i5 8gb ram nvidia 940mx :)\nwhen I hit the memory limit then I will move to the cloud
", "votes": null, "replies": [] }, { "id": 550514, "author_name": "peterwalkley", "author_url": "", "post_date": "06/11/2019 17:32:56", "content": "Intel 4790K, 16 Gig RAM and a 1070Ti.
", "votes": null, "replies": [] }, { "id": 550549, "author_name": "adilztn", "author_url": "", "post_date": "06/11/2019 18:23:14", "content": "laptop Intel Core i7-8550U 16 gb ram
", "votes": null, "replies": [] }, { "id": 550598, "author_name": "matthewmasters", "author_url": "", "post_date": "06/11/2019 19:46:21", "content": "Intel 3570K, 16GB RAM, RTX2070, running Ubuntu. I also still have a bunch of free credit for Google Cloud so I used this to generate some graph features. I've given up on competing though.
", "votes": null, "replies": [] }, { "id": 550683, "author_name": "abendorf", "author_url": "", "post_date": "06/11/2019 23:20:03", "content": "9600k, 32 GB RAM, RTX2060
", "votes": null, "replies": [] }, { "id": 550804, "author_name": "root1337x86", "author_url": "", "post_date": "06/12/2019 03:32:03", "content": "https://www.youtube.com/watch?v=_QgexYD-tZg
\n\nI7 6700K,\n32 GB DDR4\n1080 TI NVIDIA\nM.2 SSD
\n\nRUNNING: MAC OSX HIGH SIERRA (YES, IM A HACKINTOSHER XD)
", "votes": null, "replies": [ { "id": 551184, "author_name": "mindtrinket", "author_url": "", "post_date": "06/12/2019 12:33:34", "content": "I gave up on my Hackintosh... I just remote in from my macbook :)
", "votes": null, "replies": [] }, { "id": 552617, "author_name": "root1337", "author_url": "", "post_date": "06/14/2019 07:59:27", "content": "", "votes": null, "replies": [] }, { "id": 554799, "author_name": "root1337x86", "author_url": "", "post_date": "06/18/2019 03:11:44", "content": "yeah stick with ubuntu :)
", "votes": null, "replies": [] } ] }, { "id": 550939, "author_name": "venkat555", "author_url": "", "post_date": "06/12/2019 06:43:14", "content": "i tried setting up a environment on floydhub, but i am not able to get the gpu support going \nLightGBMError: GPU Tree Learner was not enabled in this build.\nPlease recompile with CMake option -DUSE_GPU=1
\n\nin the long run, having your own rig pays off
", "votes": null, "replies": [] }, { "id": 551112, "author_name": "borisdee", "author_url": "", "post_date": "06/12/2019 11:02:03", "content": "Intel Xeon E5-2637 @ 3.50GHz (16 cores)\n128 GB RAM\nNVidia Quadro M4000
", "votes": null, "replies": [] }, { "id": 551185, "author_name": "mindtrinket", "author_url": "", "post_date": "06/12/2019 12:33:39", "content": "Ubuntu 18.04\nIntel(R) Core(TM) i9-9820X CPU Overclocked to 4.1 GHz\n64 GB RAM\nRTX 2080ti
", "votes": null, "replies": [] }, { "id": 551233, "author_name": "scirpus", "author_url": "", "post_date": "06/12/2019 13:44:50", "content": "Ubuntu 18.04\nThread Ripper\n128GB RAM\nTitan Xp x2\nNVME 1Gb
", "votes": null, "replies": [ { "id": 556797, "author_name": "kingychiu", "author_url": "", "post_date": "06/20/2019 17:12:40", "content": "Haha, I am also using AMD Thread Ripper, I am using 1950X.
", "votes": null, "replies": [] }, { "id": 566686, "author_name": "barnwellguy", "author_url": "", "post_date": "07/02/2019 13:09:44", "content": "Hi, @scirpus , @kingychiu , and @taniaj !
\n\nCould you share some of the specifications of your RAM and motherboard? 😆
\n\nHope this rather technical question does not bother you guys. I am building a computer also with Threadripper (1950x) now and plan to get 128 GB RAM. But I am a little confused by the CPU's memory specifications. Most forum discussions are for smaller RAM.
\n\nAccording to AMD website (https://www.amd.com/en/products/cpu/amd-ryzen-threadripper-2950x)\nSystem Memory Specification : 2933MHz\nSystem Memory Type : DDR4\nMemory Channels : 4
\n\nBut the more RAM slots we use, it seems the lower frequency it can achieve. According to \nhttps://en.wikichip.org/wiki/amd/ryzen_threadripper/2950x
\n\n
I guess typically 128GB = 16GB * 8 and then it will be one of those \"8 of 8\" in the table above. Then, the operating frequency 2133MHz/1866MHz is much lower than the official website 2933MHz.
\n\nI am not sure what type of memory operating frequency to buy now. So, I think perhaps I can follow what you guys are using :)
\n\nThanks a lot!
", "votes": null, "replies": [] }, { "id": 566695, "author_name": "kingychiu", "author_url": "", "post_date": "07/02/2019 13:30:36", "content": "You will need 8 Ram slots. Here is how I managed to get 128GB Ram with 1950X. I built it during black friday 2018 on new egg :) (Sadly there weren't many discounts on GPUs)
\n\n
Board: GIGABYTE X399 AORUS XTREME\nRAM: Corsair DIMM 32GB DDR4-3000 Kit (x4)
", "votes": null, "replies": [] }, { "id": 566983, "author_name": "barnwellguy", "author_url": "", "post_date": "07/02/2019 21:59:20", "content": "Thanks, @kingychiu @taniaj !
\n\nI feel much more comfortable to make the purchase now!
\n\nBack Friday is good! 4th July is coming, but did not see much discount yet 🙌
", "votes": null, "replies": [] }, { "id": 567028, "author_name": "kingychiu", "author_url": "", "post_date": "07/03/2019 01:06:03", "content": "Make sure you spend money also on cooling if you will keep 100% CPU usage or even overclocking.
", "votes": null, "replies": [] }, { "id": 573170, "author_name": "suchitnasre", "author_url": "", "post_date": "07/11/2019 22:03:10", "content": "@kingychiu : what was the cost for each components on black friday sales??
", "votes": null, "replies": [] } ] }, { "id": 551405, "author_name": "abhinand05", "author_url": "", "post_date": "06/12/2019 17:17:57", "content": "Well, after reading through all this I have a feeling that my hardware is from the stone age.
", "votes": null, "replies": [ { "id": 553125, "author_name": "tobikaggle", "author_url": "", "post_date": "06/15/2019 06:30:12", "content": "@abhinand05 lol
\n\nBut going through older Kaggle competitions people used dual cores or less. Its not always about brute-forcing and massive compute power, rather some clever and efficient coding. It is simply insane how people can come up with 10 liners, when others have to program an elaborate 1000 lines of code.
\n\nSee the solutions to the Google code challenge archives, its crazy:\nhttps://codingcompetitions.withgoogle.com/codejam/archive
\n\nYes, its nice to have a 100k GPU cluster and 1000 CPU cores, but nothing beats ingenuity. Kaggle kernel only challenges could equalize the field, basically if thousands of Kagglers sit in a single room. But then we have college entry exam situations. Not sure thats a lot of fun.
", "votes": null, "replies": [] }, { "id": 553144, "author_name": "glimmung", "author_url": "", "post_date": "06/15/2019 07:23:31", "content": "Forcing Kagglers to use Python is not right. Get C# into kernels? (Oh, on topic: CPU 10 GHz (4 x 2.5), GPU: 288 GHz (480 x .6) and memory: 8 GB)
", "votes": null, "replies": [] }, { "id": 555245, "author_name": "stajh05", "author_url": "", "post_date": "06/18/2019 16:15:36", "content": "Mine is too; 7 years old. I am currently not doing great in this contest but I did well in the earthquake one. Just goes to show you can compete without fancy hardware.
", "votes": null, "replies": [] } ] }, { "id": 552787, "author_name": "venheads", "author_url": "", "post_date": "06/14/2019 14:18:16", "content": "2 machines each \n28 Xenon Cores with 56 threads\n256 GB Ram
\n\n1 DGX-1\n512 GB RAM\n8 Tesla V 100 32Gb each\nDual 20-Core Intel Xeon E5-2698 v4 2.2 GHz
", "votes": null, "replies": [ { "id": 552809, "author_name": "qiaojian", "author_url": "", "post_date": "06/14/2019 15:00:15", "content": "Wow, our company is so rich!!!!
", "votes": null, "replies": [] }, { "id": 553153, "author_name": "darbin", "author_url": "", "post_date": "06/15/2019 08:00:18", "content": "Wow
", "votes": null, "replies": [] }, { "id": 554751, "author_name": "leekltw1", "author_url": "", "post_date": "06/18/2019 01:41:55", "content": "just wow... what a beast...
", "votes": null, "replies": [] }, { "id": 558732, "author_name": "stocks", "author_url": "", "post_date": "06/22/2019 20:24:16", "content": "I'm gonna need a bigger boat...
", "votes": null, "replies": [] } ] }, { "id": 552824, "author_name": "eplistical", "author_url": "", "post_date": "06/14/2019 15:53:58", "content": "Dell xps13\nManjaro 18\n4 cores Intel i5-7200U \n8GB RAM\n128G SSD\nNo GPU
", "votes": null, "replies": [] }, { "id": 553159, "author_name": "scirpus", "author_url": "", "post_date": "06/15/2019 08:13:37", "content": "Don't forget the metric suggests to can work on each type individually so this will reduce the problem space by 8
", "votes": null, "replies": [ { "id": 553176, "author_name": "artgor", "author_url": "", "post_date": "06/15/2019 09:11:06", "content": "Not exactly, considering that different types have different amount of samples.
", "votes": null, "replies": [] }, { "id": 553200, "author_name": "scirpus", "author_url": "", "post_date": "06/15/2019 09:53:15", "content": "True but you can do what you did - calculate fc as a whole then feed that into individual models
", "votes": null, "replies": [] } ] }, { "id": 554265, "author_name": "pierscasimir", "author_url": "", "post_date": "06/17/2019 07:52:37", "content": "Standard iMac and MacbookAir
", "votes": null, "replies": [] }, { "id": 554575, "author_name": "tanmaypandey", "author_url": "", "post_date": "06/17/2019 17:18:13", "content": "I am getting some serious anxiety level lol. I bet anyone here to beat my specs:\nCompaq 610\nUbuntu 18.04\n2 GB RAM\nIntel Core 2 Duo
", "votes": null, "replies": [ { "id": 554585, "author_name": "scirpus", "author_url": "", "post_date": "06/17/2019 17:34:27", "content": "You should try using your phone ;)
", "votes": null, "replies": [] }, { "id": 554794, "author_name": "akhileshrai", "author_url": "", "post_date": "06/18/2019 03:04:39", "content": "Haha, this has changed my perspective on the kind of specs other Kagglers are using, I thought it would be closer to Alienware, Predator or MSI. I use a Macbook Pro.
", "votes": null, "replies": [] }, { "id": 554867, "author_name": "silverstone1903", "author_url": "", "post_date": "06/18/2019 05:56:55", "content": "You should try using abacus 😆
", "votes": null, "replies": [] }, { "id": 556806, "author_name": "returnofsputnik", "author_url": "", "post_date": "06/20/2019 17:23:30", "content": "I think we tie ;) Chromeos, 4 GB RAM, intel celeron. but i always use kernels
", "votes": null, "replies": [] }, { "id": 571231, "author_name": "cpmpml", "author_url": "", "post_date": "07/09/2019 10:49:05", "content": "\n\n\nI use a Macbook Pro.
\n
I fried a macbook pro with Kaggle. Sure, it was a mid 2015 and it is eligible in Apple recent recall. Batteries just blew out of proportion because of heat.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "550094": "I am wondering a lot of folks would be using their own infrastructures for this competition. How does the hardware specs look like?\n\nRAM? SSD? GPU? Cloud instance? TPU? ...........\n", "550135": "Xenon box with 64GB +64GB swap, Ubuntu Linux 18.04, 2xNVidia RTX2080ti, 500GB SSD + 4TB HDD, hybrid liquid cooling, Intel i9 with 10 cores/20 threads. For messing around Lenovo T460, 12GB Ram, no GPU, Ubuntu Linux.", "550138": "I have a home PC with i7-8700K, 64 Gb RAM and 1 1080ti.", "550140": "Threadripper 2950X, RTX-2080ti, GTX-1070ti, 128Gb RAM", "550142": "👍 Wow thats a serious machine. I guess 15 grand plus.", "550171": "Nah, half that and was pre-configured for me as well. I used to use a cloud instance with GPU on Azure but the cost of that in the long term was just not worth it. Having a decent home-office setup will last me a long time and it will pay for itself (compared to cloud). My company does blockchain anyway so plenty of other ways it is useful as well, and if I get half-good at ML and start consulting I'll need the tools of the trade.", "550172": "Awesome setup!", "550185": "Have same as Andrew but with two 1080. Have tried to get lightgbm running on GPU but with no success.", "550231": "currently laptop i5 8gb ram nvidia 940mx :)\nwhen I hit the memory limit then I will move to the cloud", "550271": "Are you on windows? I have a similar setup and absolutely could not get lightgbm to build with GPU support (or at all, really). \n\nHowever, on my Linux machine building it for gpu worked like a charm.", "550514": "Intel 4790K, 16 Gig RAM and a 1070Ti.", "550549": "laptop Intel Core i7-8550U 16 gb ram", "550598": "Intel 3570K, 16GB RAM, RTX2070, running Ubuntu. I also still have a bunch of free credit for Google Cloud so I used this to generate some graph features. I've given up on competing though.", "550683": "9600k, 32 GB RAM, RTX2060", "550804": "https://www.youtube.com/watch?v=_QgexYD-tZg\n\nI7 6700K,\n32 GB DDR4\n1080 TI NVIDIA\nM.2 SSD \n\nRUNNING: MAC OSX HIGH SIERRA (YES, IM A HACKINTOSHER XD)", "550939": "i tried setting up a environment on floydhub, but i am not able to get the gpu support going \nLightGBMError: GPU Tree Learner was not enabled in this build.\nPlease recompile with CMake option -DUSE_GPU=1\n\nin the long run, having your own rig pays off", "551112": "Intel Xeon E5-2637 @ 3.50GHz (16 cores)\n128 GB RAM\nNVidia Quadro M4000", "551184": "I gave up on my Hackintosh... I just remote in from my macbook :)", "551185": "Ubuntu 18.04\nIntel(R) Core(TM) i9-9820X CPU Overclocked to 4.1 GHz\n64 GB RAM\nRTX 2080ti", "551233": "Ubuntu 18.04\nThread Ripper\n128GB RAM\nTitan Xp x2\nNVME 1Gb", "551405": "Well, after reading through all this I have a feeling that my hardware is from the stone age.", "552617": "", "552787": "2 machines each \n28 Xenon Cores with 56 threads\n256 GB Ram\n\n1 DGX-1\n512 GB RAM\n8 Tesla V 100 32Gb each\nDual 20-Core Intel Xeon E5-2698 v4 2.2 GHz", "552809": "Wow, our company is so rich!!!!", "552824": "Dell xps13\nManjaro 18\n4 cores Intel i5-7200U \n8GB RAM\n128G SSD\nNo GPU", "553125": "abhinand05 lol\n\nBut going through older Kaggle competitions people used dual cores or less. Its not always about brute-forcing and massive compute power, rather some clever and efficient coding. It is simply insane how people can come up with 10 liners, when others have to program an elaborate 1000 lines of code.\n\nSee the solutions to the Google code challenge archives, its crazy:\nhttps://codingcompetitions.withgoogle.com/codejam/archive\n\nYes, its nice to have a 100k GPU cluster and 1000 CPU cores, but nothing beats ingenuity. Kaggle kernel only challenges could equalize the field, basically if thousands of Kagglers sit in a single room. But then we have college entry exam situations. Not sure thats a lot of fun.", "553144": "Forcing Kagglers to use Python is not right. Get C# into kernels? (Oh, on topic: CPU 10 GHz (4 x 2.5), GPU: 288 GHz (480 x .6) and memory: 8 GB)", "553153": "Wow", "553159": "Don't forget the metric suggests to can work on each type individually so this will reduce the problem space by 8", "553176": "Not exactly, considering that different types have different amount of samples.", "553200": "True but you can do what you did - calculate fc as a whole then feed that into individual models", "554265": "Standard iMac and MacbookAir", "554575": "I am getting some serious anxiety level lol. I bet anyone here to beat my specs:\nCompaq 610\nUbuntu 18.04\n2 GB RAM\nIntel Core 2 Duo", "554585": "You should try using your phone ;)", "554751": "just wow... what a beast...", "554794": "Haha, this has changed my perspective on the kind of specs other Kagglers are using, I thought it would be closer to Alienware, Predator or MSI. I use a Macbook Pro.", "554799": "yeah stick with ubuntu :)", "554867": "You should try using abacus 😆", "555245": "Mine is too; 7 years old. I am currently not doing great in this contest but I did well in the earthquake one. Just goes to show you can compete without fancy hardware.", "556797": "Haha, I am also using AMD Thread Ripper, I am using 1950X.", "556806": "I think we tie ;) Chromeos, 4 GB RAM, intel celeron. but i always use kernels", "558732": "I'm gonna need a bigger boat...", "566686": "Hi, @scirpus , @kingychiu , and @taniaj !\n\nCould you share some of the specifications of your RAM and motherboard? 😆 \n\nHope this rather technical question does not bother you guys. I am building a computer also with Threadripper (1950x) now and plan to get 128 GB RAM. But I am a little confused by the CPU's memory specifications. Most forum discussions are for smaller RAM.\n\nAccording to AMD website (https://www.amd.com/en/products/cpu/amd-ryzen-threadripper-2950x)\nSystem Memory Specification : 2933MHz\nSystem Memory Type : DDR4\nMemory Channels : 4\n\nBut the more RAM slots we use, it seems the lower frequency it can achieve. According to \nhttps://en.wikichip.org/wiki/amd/ryzen_threadripper/2950x\n\n\n\nI guess typically 128GB = 16GB * 8 and then it will be one of those \"8 of 8\" in the table above. Then, the operating frequency 2133MHz/1866MHz is much lower than the official website 2933MHz.\n\nI am not sure what type of memory operating frequency to buy now. So, I think perhaps I can follow what you guys are using :)\n\nThanks a lot!", "566695": "You will need 8 Ram slots. Here is how I managed to get 128GB Ram with 1950X. I built it during black friday 2018 on new egg :) (Sadly there weren't many discounts on GPUs)\n\n", "566701": "Board: GIGABYTE X399 AORUS XTREME\nRAM: Corsair DIMM 32GB DDR4-3000 Kit (x4)", "566983": "Thanks, @kingychiu @taniaj !\n\nI feel much more comfortable to make the purchase now!\n\nBack Friday is good! 4th July is coming, but did not see much discount yet 🙌", "567028": "Make sure you spend money also on cooling if you will keep 100% CPU usage or even overclocking.", "571231": "> I use a Macbook Pro.\n\nI fried a macbook pro with Kaggle. Sure, it was a mid 2015 and it is eligible in Apple recent recall. Batteries just blew out of proportion because of heat.", "571906": "I have the same question with you.I use Window10.But XGBoost and Tensorflow is workable running on GPU .", "573170": "kingychiu : what was the cost for each components on black friday sales??" }, "source": "meta" }