Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Ask HN: Any affordable deep learning cloud service for practicing/experimenting?
27 points by febin on April 17, 2018 | hide | past | favorite | 16 comments


Crestle (https://www.crestle.com/): + super quick to setup + free trial no CC required o underlying GPUs are ok - more expensive than others

Google Colab (https://colab.research.google.com/): + Free + Works as a plugin within Google Drive - Very limited hardware - Integration with data storage could be easier

I'm going to try Gradient as well. Paperspace has very competitive pricing and their new service seems to hit the nail in terms of ease of use vs control of settings. Any chance @dkobran has a coupon code to take it for a spin? :)


There are referral codes floating all over the internet ;)



We just launched Vectordash! (http://vectordash.com) Someone mentioned us earlier too - the fundamental idea behind it is that gradient descent should not be expensive and compute costs should never be the limiting factor in deep learning. Our GPU instances cost 5x less than AWS because we want everyone to be able to make progress in deep learning regardless of their budget.


I’ve heard Floydhub is good too. https://www.floydhub.com/pricing


Recently there was a hacker news post about using gpus in mining rigs as gpu instances to perform deep learning.

https://vectordash.com/

It seemed like a really cool idea. And the instances were pretty cheap.

Note: I am not sure if they solved the "verification of solution" problem yet.


Paperspace Cofounder here. We just launched Gradient, a really simple and affordable DL platform: https://paperspace.com/gradient Feel free to ping me with any questions.


I really liked Paperspace Core.

Had to cancel my subscription because my daughter was born and I knew I won't have free nights to hack until 2019.


Here’s someone who knows exactly what he’s getting into.

Took me months to realize that I wouldn’t get back to fun hacking for years. (Thought I’d only be out for a couple months)


>10 GPU Jobs Limit (1 concurrent)

What does that actually mean in a PAYG plan?


Just that you’re limited to the number of saved and running experiments (Jobs) on the free plan. For most research/non-production environments, this is sufficient. Hope that helps.


>10 GPU Jobs Limit (1 concurrent)

I also found this part of Paperspace's pitch confusing, to be honest, to the point that I worried it was some tricky restriction, and bailed out midway through the sign up process.

You have jobs, runs, three types of storage, various machine types, with fixed monthly and per minute charges. The main pricing page doesn't say what these terms mean. This all makes sense to you guys (as it is your company). But an average prospective customer will give up and go elsewhere.

Edit: I just created a paperspace.com account, logged in, and again couldn't figure out how your pricing for Gradient works. The missing info is on a help centre page that explains the three types of storage (workspace, artifact and persistent), but it assumes I know what a job is. If I had more time, I'd help you rewrite these docs into a single integrated guide!


I mostly use floydhub for its ease of use.

Colab from google lets you train over GPUs for upto 12 hours for free, However it takes a while to setting things up esp pytorch and I'm not a fan of the UI.


How about Google Cloud‘s Cloud Machine Learning Engine? They even provide 300$ free credits on registration.


This, currently using them. Works great. They do request you to deposit around 75 dollars to make sure you aren't a fraud that is going to mine 300 dollars worth of bitcoins. The 75 dollars aren't gone, and to be honest its worth the money !


Colab if you're beginner




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: