Cheapest (0$) Deep Learning Hardware Options | 2021

Key Takeaways

okay so you started playing with uh deep learning and you figured out that neural networks take take up a lot of vram uh they are really compute intensive and you'll need some expensive hardware specifically gpus in order to do anything meaningful so in this video i'm going to break it down for you and start with the uh cheapest options like zero dollars uh like free options first and then in the second part of the video i'll give you some recommendations about the paid options also so without further ado let's uh start with the first option okay so the first two options you should consider uh are both free and they are google collab and kaggle so you probably heard of both of them actually they both belong to google google collab is awesome because it gives you gpus and also tpus for free and the same goes for kaggle which gives you p hundred gpus which are actually a bit better than collabs because colab gives you k80s there are a lot of tutorials how to get started and i won't get into that part i'll just give you all the options you have there are obviously some trade-offs when using google collab and using kaggle and the main ones is that they disconnect after a certain period so for collab you can expect around 12 hours before it kind of just disconnects your runtime and kaggle gives you around nine hours of uh free usage now if your gpu goes idle it will basically stop your your runs and there's one more thing you should know and that's that uh there is no guarantee that colab will give you gpus or tpus they have a certain amount of gpus per region so if if people in your region are using collab a lot either for deep learning or maybe mining then you probably won't get a gpu and that sucks okay for the second uh part of this video i'm going to tell you what your next three options are and that's cloud and i hear you say wait cloud but the thing is most of the big traditional cloud providers such as azure such as aws google cloud platform or gcp for short they all give you like 200 to 300 dollars for a month for free and you can kind of use that so azure gives you 200 bucks for a month also uh ibm gives you 200 uh for a month gcp gives you three hundred dollars for a month and finally uh alibaba gives you three hundred dollars you also have aws has some free program but i'm not sure if they have the same thing as these ones i mentioned i think they only offer cpus so doing this you can basically uh have a like uh you could have uh cloud gpus like k80s uh for four or five months if you just switch between these providers and it's kind of annoying to kind of to switch all the time but you basically need one day to set it up and get started and then you have it free for a month and then you can just switch to the other cloud provider doing this you'll acquire important skill and that's using different cloud providers later if you want to maybe continue using some of them you can just pay and keep using the provider so aside from these traditional cloud providers you also have a machine learning specific cloud providers such as spell which i've used and which is really super easy to use much easier than some of these more traditional cloud providers there is also paper space there is floyd hub there are a bunch of options there so spell gives you 10 bucks for free and you can check for the others a link a bunch of useful resources down in the description so the thing with these ml specific cloud providers is that they are all based off of like traditional providers so they use as a backbone to use either aws or azure or gcp and so forth and that was basically it for the free options so you either use colab or kaggle for free forever uh but obviously you it's slower it sometimes disconnects your runtime so there are some cons there or you simply use you switch from cloud provider to cloud provider you you get a lot of skill by doing that and you get some awesome gpu clusters for free so those were the three options now let's jump into some of the paid options which you may consider if you're doing deep blurring really seriously so there are basically two things you could do here you could either just buy some hardware off the shelf or you could build your own custom pc you can basically build a deep learning pc for less than thousand bucks and it could be really good i'll link a couple of those videos for building a budget deep learning pc down in the description if you have more money you can build obviously better deep learning pcs also called those deep learning rigs so basically if you can afford two or three or four thousand dollars you can build a really a beast of a pc and the reason why you you might want to do this is if you're really doing some serious deep learning then you'll either have to pay some cloud-based solution or you'll build your custom pc and there are a bunch of resources out there that show that it's much more cost effective to actually build your own deep learning rig and i'll also link some of the resources down in the description uh helping you build the rig for yourself so at the end it all depends on your preference if you if you don't have any budget whatsoever and you want to start with machine learning then go with kegel or go with google collab uh if you do have some money but you don't want to lose any time you just want something quick just buy some off-the-shelf pc or a deep learning rig if you do have some money and you also do have some time to spend you can the best thing to do is to build your your own deep learning rig and i think that the worst option if you're an individual playing with deep learning is to be paying the cloud unless you are a startup and you create some kind of a multi-year contract so that's it for this video hope you found it useful so you have all of those options out there i'll link a bunch of useful resources down in the description if you wanna maybe investigate a bit deeper but that's like the the high level overview of the of the current hardware uh deep learning hardware uh landscape so i'd like to know which options out of these uh did you personally use and also which one works the best for you uh in your current situation if you think i've missed something uh feel free to comment uh down in the comment section and i'll get back to you if you found this video useful consider subscribing and hit that like button to get notified when i upload a new video until next time keep learning deep [Music] you