GPU server on cloud


(Sankar R) #1

Came across a GPU server on cloud for ml. It says approximately $200/mo or 0.67/hr https://www.paperspace.com/ml Has anyone had any experience with it.


(Samuel Ekpe) #2

i have seen it, but i havent tried it, maybe we can


(jerry liu) #3

I just created a GPU instance on paperspace. 8GB GPU for $0.40/hr. Setup was totally painless and came preconfigured with Keras 2.0 and Tensorflow 1.1

Anybody have experience with their billing?


(Sankar R) #4

Hi Jerry, Can you give me some feedback regarding Paperspace. How was/is your experience? Are you still using it? Thanks Sankar


(jerry liu) #5

I’m still using it, and I’m happy to report that they invoice per minute billing. They also have a setting to switch off the machine after X hours, in case I forget.

I think the GPU is less powerful, but I find it much easier to set up and get going, no messing about with AWS. Probably the only thing I configured was to turn off password SSH.

If it helps, heres a referral code: OM6JAO


(Sankar R) #6

Thank you very much.


#7

Also, have a look at:
https://www.hetzner.com/dedicated-rootserver/ex51-ssd-gpu?country=ot
99eur/month (+99eur setup fee)

Specs are great (i7-6700 Quad-Core, GTX1080, 64GB RAM, 2 x 500 GB SATA) but I haven’t used them yet.


(Sankar R) #8

Hi Jerry, Took the plunge and joined paperspace.
Had to select linux ml-in-a-box to the get the GPU+ machine.

Now that I got the machine, i did not find anaconda installed.
Did you install anaconda or are you not using jupyter notebooks.

Thanks


(jerry liu) #9

Sankar - I think anaconda is fairly easy to install? No I don’t use jupyter on cloud GPU… I typically experiment on local machine and then just run a queue of experiments on cloud GPU when im ready.

HTH


(Thomas van den Berg) #10

I was using AWS before, as set up by the course’s setup scripts.

I’m now trying out Paperspace and found it quite a bit easier to get into compared to AWS.
I installed conda through miniconda and then used that to install jupyter. Took me about an hour to figure that out.

Ideally I’d love to get all my setup into a Docker image or similar and be free to switch platforms at whim.

However that requires nvidia-docker and that gave me some headaches so I put it off for now.