[{"data":1,"prerenderedAt":519},["ShallowReactive",2],{"project-26492931":3},{"id":4,"name":5,"line":6,"line_source":7,"title":8,"url":9,"domain":5,"maker":10,"posted_at":11,"year":12,"points":13,"comments":14,"rank_in_year":15,"shown_in_year":16,"top":17,"tier":18,"category":19,"subcategory":20,"format":21,"keywords":22,"reception":27,"sentiment":28,"stance_mix":29,"scored_comments":34,"aspects":35,"strip":47,"post_text":48,"what":49,"said":83,"maker_in_thread":229,"shape":240,"grid":245,"attempts":480,"same_week":481,"maker_shows":37,"maker_hidden":511,"reader_version":512,"card":513},"26492931","gpu.land","Cloud GPUs for Deep Learning – At 1\u002F3 the Cost of AWS\u002FGCP","title","Show HN: Cloud GPUs for Deep Learning – At 1\u002F3 the Cost of AWS\u002FGCP","https:\u002F\u002Fgpu.land\u002F","ilmoi",1615996605,2021,168,93,286,2533,12,"notable","infrastructure","cloud gpu rental","web_app",[23,24,25,26],"cloud gpu","deep learning","tesla v100","gpu rental","mixed",0.024,{"positive":30,"neutral":31,"mixed":32,"critical":33},13,23,3,11,50,{"idea":36,"performance":39,"pricing":40,"execution":42,"privacy":44,"design":45,"docs":46},{"positive":37,"critical":38},1,0,{"positive":38,"critical":37},{"positive":32,"critical":41},4,{"positive":37,"critical":43},5,{"positive":38,"critical":38},{"positive":37,"critical":38},{"positive":38,"critical":38},[32,32,43,37,38,32,43,38,32,43,32,43,32,32,32,32,43,37,38,38,37,32,38,32,32,32,37,32,37,38,37,32,32,37,38,32,43,43,43,43],null,{"problem":50,"audience":57,"pricing":64,"motivation":75,"time_to_build":48,"stack":79,"openness":80,"stage":81,"team_size":82},{"value":51,"source":52,"evidence":53},"Cloud GPUs from AWS\u002FGCP are expensive and slow to set up for deep learning","maker",[54],{"id":55,"by":10,"text":56},"26492960","I’m a self taught ML engineer, and when I was starting on my ML journey I was incredibly frustrated by cloud services like AWS\u002FGCP. a)they were super expensive, b)it took me…",{"value":58,"source":52,"evidence":59},"ML engineers, students, researchers and hobbyists needing GPUs",[60,61],{"id":55,"by":10,"text":56},{"id":62,"by":10,"text":63},"26500111","Great questions. 1) There's a finite capacity in the high 10s of GPUs. Currently the service is utilized at \u003C10% capacity, so there's a lot of room to grow before we run out. If…",{"value":65,"source":52,"evidence":66},"$0.99\u002Fhr per Tesla V100, $0.02\u002FGB\u002Fmonth storage, top-up credit",[67,68,72],{"id":55,"by":10,"text":56},{"id":69,"by":70,"text":71},"26495308","mastermojo","Very cool. Is there a way to persist disks at a cost and attach\u002Funattach Voltas for training? EDIT: found in the FAQ: Compute: $0.99\u002Fhr \u002F 1x Tesla V100 (running instance only)…",{"id":73,"by":10,"text":74},"26495546","Yep - most people stop instances and come back whenever they're ready to train again. No charges for GPUs while the instance is stopped. If you have a say 200GB harddrive you're…",{"value":76,"source":52,"evidence":77},"Maker frustrated with expensive, slow cloud GPU setup as a self-taught ML engineer; wants to give back",[78],{"id":55,"by":10,"text":56},[],"unknown","launched","solo",{"praise":84,"objections":105,"requests":148,"questions":177,"alternatives":200},[85,94,98],{"value":86,"source":87,"evidence":88},"Much cheaper than hyperscale clouds","commenter",[89,90],{"id":55,"by":10,"text":56},{"id":91,"by":92,"text":93},"26494014","dindresto","How does the Tesla V100 compare to the Tesla P100, which Scaleway offers at a similar price of €1\u002Fhour? https:\u002F\u002Fwww.scaleway.com\u002Fen\u002Fgpu-instances\u002F",{"value":95,"source":52,"evidence":96},"Simple setup and pre-configured environments",[97],{"id":55,"by":10,"text":56},{"value":99,"source":87,"evidence":100},"Top-up feature prevents accidental overspending",[101],{"id":102,"by":103,"text":104},"26494535","ashish01","Nice! I really like the top up feature. Really gives me peace of mind that I will not blow up my budget accidentally. I have been using https:\u002F\u002Fdatacrunch.io\u002F which has almost…",[106,117,127,138],{"value":107,"source":87,"evidence":108},"Reliability and uptime compared to hyperscalers unclear",[109,113],{"id":110,"by":111,"text":112},"26494778","sacheendra","This I highly unlikely. Hyperscalers like AWS have custom power delivery, rack organization, redundant networking, etc. which make their instances reliable. Long running jobs…",{"id":114,"by":115,"text":116},"26495137","tombh","These seems very similarly priced to AWS Spot Instances. But you're guaranteeing uptime?",{"value":118,"source":87,"evidence":119},"Finite capacity, only up to 16 GPUs per account",[120,124],{"id":121,"by":122,"text":123},"26497071","iujjkfjdkkdkf","Sorry if I missed it in the FAQ, do you have finite capacity or do you have an agreement with the data center where you can bring on more gpus with demand. My worry would be to…",{"id":125,"by":10,"text":126},"26496079","You can! All machines within a single account are inside a private VLAN, which means they all talk to each other (out of the box, no setup required), but nobody else on gpu.land…",{"value":128,"source":87,"evidence":129},"Site scrolling and loading broken in some browsers and adblockers",[130,134],{"id":131,"by":132,"text":133},"26494555","ffpip","FYI, the site shows nothing until the scripts from stripe.com load. I know that that these breakages might not happen to all users, but it shouldn't be completely broken until…",{"id":135,"by":136,"text":137},"26499275","Reubend","Hey there, just wanted to share that for me (using Opera browser with ad block) scrolling is broken on your landing page.",{"value":139,"source":87,"evidence":140},"Only Linux hosts, no containers support",[141,145],{"id":142,"by":143,"text":144},"26495367","paul_milovanov","this is great, except apparently there's no support for containers? is that on the roadmap? thanks!",{"id":146,"by":10,"text":147},"26500140","Not externally (in the same way that vast.ai does it), but you could always boot the instance -> ssh in -> manually run nvidia-docker. Added to feature requests!",[149,158,169],{"value":150,"source":87,"evidence":151},"Programmatic access and docker\u002Fenvironment deployment",[152,155],{"id":153,"by":122,"text":154},"26497394","One more comment \u002F suggestion: I watched the video and saw there are lots of pre-configured environments. I see this a lot, but to be honest this is not to helpful for me because…",{"id":156,"by":10,"text":157},"26500123","Both are valid points. And no, there's currently no programmatic access to machines - but that's high on the roadmap. Regarding environments, it's funny, when I was starting out…",{"value":159,"source":87,"evidence":160},"More persistent storage options and storage details",[161,165],{"id":162,"by":163,"text":164},"26545078","bayesianasian","Is there an option to request for more persistent storage? From the website, I inferred that each 8x GPU machine comes with 2TB SSD storage? Would I be able to request for…",{"id":166,"by":167,"text":168},"26495231","perennate","The website design put me off but the FAQ makes it clear that it is quite professional. They have information about physical security, GDPR compliance, and VLAN. Wish there was…",{"value":170,"source":87,"evidence":171},"Support for multi-node training with more than 16 GPUs",[172,176],{"id":173,"by":174,"text":175},"26496021","typon","Hey ilmoi, Do you support multi-node training? Particularly, can I reserve for example a 64-GPU instance via 8 nodes and perform distributed training?",{"id":125,"by":10,"text":126},[178,193],{"value":179,"source":87,"evidence":180},"How does it compare to spot instances and vast.ai",[181,185,189],{"id":182,"by":183,"text":184},"26494795","usmannk","This is seriously cool. I want to mention that on AWS you can reliably get V100's for $0.918\u002Fhr (I do this all the time) by using spot instances. The spot price hasn't changed by…",{"id":186,"by":187,"text":188},"26498090","tehsauce","For anyone looking for cheap gpus (quite a bit cheaper than this even), I highly recommend the marketplace on vast.ai I’ve spent hundreds of dollars there and will spend many more.",{"id":190,"by":191,"text":192},"26498766","bitobserver","> At 1\u002F3 the Cost of AWS\u002FGCP For on-demand this is true, but spot GPU instances can have competitive pricing. At this moment, a p3.2xlarge instance (1 Tesla v100) is 92 cents\u002Fhr…",{"value":194,"source":87,"evidence":195},"Data persistence and checkpointing reliability",[196],{"id":197,"by":198,"text":199},"26494528","liuliu","Thank you! This looks indeed cheap. Can you share more stories on data persistence \u002F checkpointing? If I have a job that requires 8 V100 with 3 days, what types of reliability am…",[201,206,213,220],{"value":202,"source":87,"evidence":203},"AWS and GCP",[204,205],{"id":55,"by":10,"text":56},{"id":182,"by":183,"text":184},{"value":207,"source":87,"evidence":208},"vast.ai",[209,210],{"id":186,"by":187,"text":188},{"id":211,"by":10,"text":212},"26495252","You are correct that vast.ai is pretty close. The biggest difference is probably security \u002F guaranteed uptime. With vast you're getting what it says on the tin - a machine from a…",{"value":214,"source":87,"evidence":215},"Google Colab",[216],{"id":217,"by":218,"text":219},"26497681","mraza007","Hey I loved the product you made. As a student who is always concerned about the spending on cloud services as they can be pricey but I’m just curious I have used google Colab in…",{"value":221,"source":87,"evidence":222},"Scaleway, Hetzner, LeaderGPU, Datacrunch",[223,224,228],{"id":91,"by":92,"text":93},{"id":225,"by":226,"text":227},"26495263","etaioinshrdlu","Can you use GeForce cards and then make us pinky swear to only use it for blockchain processing applications? This is what LeaderGPU does (https:\u002F\u002Fwww.leadergpu.com\u002F#chose-best),…",{"id":102,"by":103,"text":104},{"comments":230,"answered":231,"key":232},29,15,[233,234,237],{"id":55,"text":56},{"id":235,"text":236},"26493153","thanks for reading the FAQ! Most people don't seem to bother:)",{"id":238,"text":239},"26495503","CUDA develpment is totally fine. The service was designed with DL in mind, but that shouldn't stop CUDA dev in any way.",{"commenters":241,"max_depth":242,"maker_replies":230,"first_comment_mins":243,"lifespan_hours":244},49,7,2,216,[246,248,252,254,257,258,262,264,266,268,272,275,277,280,281,283,286,289,291,292,294,297,299,301,302,304,307,311,314,318,320,322,324,327,328,329,330,331,334,338,342,346,347,349,351,354,357,362,365,367,370,372,374,377,379,383,386,388,391,394,396,398,401,404,405,408,410,413,415,417,419,421,423,427,430,432,433,435,438,440,442,444,446,448,451,452,454,458,463,466,470,474,478],[55,38,38,247,10,37],"project\u002Fpositive\u002Fpricing",[249,38,32,250,251,38],"26493075","project\u002Fneutral\u002Fnone","chovybizzass",[235,38,38,253,10,37],"maker\u002Fneutral\u002Fdocs",[255,38,32,250,256,38],"26495431","Sanzig",[238,38,38,250,10,37],[259,38,43,260,261,38],"26494209","project\u002Fpositive\u002Fidea","ogiberstein",[197,38,32,263,198,38],"project\u002Fneutral\u002Fperformance",[265,38,38,263,10,37],"26494541",[110,38,37,267,111,38],"project\u002Fcritical\u002Fperformance",[269,38,38,270,271,38],"26499117","commenter\u002Fneutral\u002Fperformance","shrubble",[273,38,38,270,274,38],"26499310","nixgeek",[276,38,38,270,271,38],"26505919",[278,38,32,263,279,38],"26495099","joecot",[102,38,43,247,103,38],[225,38,32,282,226,38],"project\u002Fneutral\u002Fpricing",[284,38,43,247,285,38],"26495301","ghoomketu",[287,38,38,288,167,38],"26495335","other\u002Fneutral\u002Fnone",[290,38,38,288,285,38],"26500343",[173,38,32,263,174,38],[125,38,38,293,10,37],"project\u002Fneutral\u002Fexecution",[295,38,32,293,296,38],"26496929","teruakohatu",[298,38,38,282,10,37],"26500075",[300,38,43,247,296,38],"26500310",[217,38,32,282,218,38],[303,38,38,250,10,37],"26500058",[305,38,43,306,218,38],"26504998","maker\u002Fpositive\u002Fnone",[308,38,32,309,310,38],"26500524","maker\u002Fneutral\u002Fprivacy","benohear",[312,38,38,313,10,37],"26510499","maker\u002Fneutral\u002Fnone",[315,38,32,316,317,38],"26502092","project\u002Fneutral\u002Fprivacy","taf2",[319,38,38,316,10,37],"26521081",[182,37,32,321,183,38],"project\u002Fmixed\u002Fpricing",[323,37,38,282,10,37],"26495290",[325,37,32,326,183,38],"26497220","maker\u002Fneutral\u002Fexecution",[121,243,32,263,122,38],[62,243,38,250,10,37],[153,32,32,293,122,38],[156,32,38,293,10,37],[332,32,43,333,122,38],"26502307","project\u002Fpositive\u002Fexecution",[335,41,43,336,337,38],"26494575","project\u002Fpositive\u002Fdesign","maremmano",[339,43,37,340,341,38],"26493832","project\u002Fcritical\u002Fpricing","37ef_ced3",[343,43,38,344,345,38],"26495073","commenter\u002Fneutral\u002Fnone","rckoepke",[211,43,38,250,10,37],[348,43,38,344,167,38],"26495270",[350,43,38,344,341,38],"26495349",[352,43,38,288,353,38],"26498472","sharemywin",[355,43,38,356,92,38],"26500419","other\u002Fneutral\u002Fperformance",[358,359,37,360,361,38],"26493847",6,"project\u002Fcritical\u002Fnone","DIVx0",[363,359,38,364,10,37],"26493922","project\u002Fneutral\u002Fidea",[166,242,32,366,167,38],"project\u002Fmixed\u002Fdocs",[368,242,38,369,10,37],"26496246","project\u002Fneutral\u002Fdocs",[91,371,32,263,92,38],8,[373,371,38,263,10,37],"26494091",[375,371,38,344,376,38],"26494187","knrz",[378,371,38,313,10,37],"26494511",[380,381,32,250,382,38],"26494159",9,"DougN7",[384,381,32,282,385,38],"26494399","imhoguy",[387,381,38,282,10,37],"26494527",[389,381,38,344,390,38],"26495196","ska",[392,381,32,250,393,38],"26494687","jsnell",[395,381,38,313,10,37],"26495342",[190,397,37,340,191,38],10,[399,33,37,340,400,38],"26498231","lostmsu",[402,17,43,306,403,38],"26493326","akrymski",[162,30,32,250,163,38],[131,406,37,407,132,38],14,"project\u002Fcritical\u002Fexecution",[409,406,38,326,10,37],"26494627",[411,406,37,407,412,38],"26495473","jandrese",[414,406,38,326,10,37],"26495555",[416,406,38,326,10,37],"26500289",[418,406,32,293,412,38],"26503330",[420,406,37,407,412,38],"26506306",[422,406,37,407,132,38],"26511823",[424,231,32,425,426,38],"26494609","project\u002Fneutral\u002Fdesign","stuartbman",[428,231,38,429,10,37],"26495302","maker\u002Fneutral\u002Fdesign",[69,431,32,282,70,38],16,[73,431,38,282,10,37],[186,434,37,340,187,38],17,[436,434,38,437,10,37],"26510506","commenter\u002Fneutral\u002Fpricing",[135,439,37,407,136,38],18,[441,439,38,326,10,37],"26500127",[443,439,38,326,10,37],"26500284",[114,445,32,282,115,38],19,[447,445,32,369,167,38],"26495284",[142,449,32,450,143,38],20,"project\u002Fmixed\u002Fexecution",[146,449,38,326,10,37],[453,449,43,306,143,38],"26593874",[455,456,38,288,457,38],"26499249",21,"2Gkashmiri",[459,460,43,461,462,38],"26493805",22,"project\u002Fpositive\u002Fnone","neil1",[464,31,43,461,465,38],"26494259","artem_mazur",[467,468,43,461,469,38],"26493384",24,"samsammurphy",[471,472,43,461,473,38],"26522991",25,"seefashion",[475,476,38,288,477,38],"26494226",26,"wrongdonf",[479,476,38,313,10,37],"26494637",[],[482,487,492,497,502,507],{"id":483,"name":484,"points":485,"category":486},"26495059","Fontemon",605,"games",{"id":488,"name":489,"points":490,"category":491},"26475898","SELLFF",249,"web_saas",{"id":493,"name":494,"points":495,"category":496},"26491764","Dendron",237,"productivity",{"id":498,"name":499,"points":500,"category":501},"26453840","Critter.Camera",175,"hardware",{"id":503,"name":504,"points":505,"category":506},"26484339","Pyroscope",166,"dev_tools",{"id":508,"name":509,"points":510,"category":501},"26472366","fixmyspeakers",148,false,"v2",{"id":514,"name":5,"line":6,"ai":511,"cat":19,"year":12,"points":13,"comments":14,"rec":27,"s":515,"top":17,"kw":516,"maker":10,"domain":5,"tier":18,"attempts":38,"strip":517,"url":9,"mix":518,"praise":86,"objection":107,"sub":20},26492931,0.02,[23,24,25,26],"3351035035353333510013033313101331035555",[30,31,32,33],1794036604589]