AWS - request to increase limit to g2.2x large was rejected!


(vedshetty) #1

As the title says, my request to access the gpu instance was rejected by AWS even though I was a past customer (had a tiny bill from them a couple of months back). I admit the description I provided was short: Deep learning course - they probably did not believe me. Plus the recent DDOS attack also did not probably help.
I reopened the case with a little bit more details on why I needed the instance but curious to know if anyone else had an issue with denial or acceptance from AWS.


(yad.faeq) #2

Hey There,

This is normal, if you haven’t used the GPU instances with your current amazon account.

For security concerns against massive spam-bots to signup for free credit, there is an extra step to request for larger instances.

Usually the error is something like this:

You have requested more instances (1) than your current instance limit of 0

You can submit a request through here


(Jeremy Howard) #3

@yad.faeq - it appears that @vshets is saying that they have already requested the limit increase, but it was rejected.

I’m hoping that the USF folks will be able to help us with this - they’ve asked their AWS contacts to provide support.


(yad.faeq) #4

Apologies, I missed that first potion when reading it on the phone.

it might have to do with verified-non verified account requesting the limit increase perhaps. (I have had this issue before, but not with GPU instances)


(vedshetty) #5

Yeah … not sure what is going on … no tech support from AWS is assigned to my case yet (24hrs have passed since I reopened my case - guess it is the weekend). Hoping to get a resolution before the course starts tomorrow.


#6

My increase reques (for a EC2 p2.xlarge) t was just denied based previous account usage. I have had the account for more than a year, but up until now, have only utilized free EC2 instances. My concern with this sort of policy from Amazon is that it does not provide a clearly delineated path for those looking to scale up their AWS usage due to taking a Deep Learning MOOC like Fast.ai.


(Ramesh Sampath) #7

You can use Paperspace.com Unless you have AWS Credits, Paperspace will be cheaper and there’s a Fast.AI image ready to go there. Search in the forums for Paperspace setup. I think there’s a $15 Fast.ai paperspace credit somewhere in the forum.

Note that you will also have to apply in Paperspace requesting for GPU machines. I would suggest put details that you are doing Fast.AI course and having a Valid Credit Card on the file will help getting it approved.


#8

I have been using Crestle up until now. I tried AWS for two reasons, first I chose to treat it a part of Lesson 2 Homework. In that line, I thought that getting some extra AWS experience under my belt could not hurt. Secondly, I liked the idea of using it occasionally (couple of hours at a time), as Jeremy suggested, to train larger models.

Paper Space was the first account I tried setting up, even before Crestle. However, when I tried to create an Ubuntu instance, I was prompted to provide a reason for my Ubuntu Instance Request. I wrote that it was for purposes of taking the Fast.ai course, however I have yet to hear back from Paper Space.