NVIDIA GPU Cloud

Hi @jeremy ,

The NVIDIA GPU Cloud is open for business.

Website: https://www.nvidia.com/en-us/gpu-cloud/

Seems to run with AWS underneath.

Steeve

4 Likes

Not much to see here. NVIDIA GPU Cloud should be renamed to “Machine Learning Docker Container by NVIDIA”. It’s a container available for a AWS EC2 P3 Instance. You are charged full price for the instance.

Video on how to setup NVIDIA GPU Cloud: https://www.nvidia.com/en-us/gpu-cloud/?ncid=van-gpu-cloud

2 Likes

Hi @schumi -
I am not sure if it’s useful discussion on Part 1 V2. We already have $500 in credits from AWS. Also have other options like https://www.crestle.com/ and https://www.paperspace.com/ that have specific AMIs for Fast.AI.

So unless we have Nvidia giving us Credits or you think its better than other options we have already, let’s not create new Threads on things we most likely will not use in the course. It’s getting hard to keep up with all the new threads…just a suggestion.

I just wanted to point out that NVIDIA GPU cloud is not really a separate GPU cloud provider. They are just providing customized AMI optimised for the new Amazon P3 instances. I think AMI itself is free. Jeremy can probably tell us which AMI is the best to go with in AWS or whether there is going to be a custom FAST.AI AMI.

Cool. Thanks for the clarification. Now that makes sense to bring it up.

P3 instances are so much more expensive. :frowning: I will probably stick with G2 or P2 instances. http://www.ec2instances.info/

paperspace has the best value GPU instances at the moment, IMO.

2 Likes

To revive this thread…

has anyone taken the NVIDIA Pytorch Container and created a new container with anaconda installed and all the libraries used in part1 such that it can be shared for those running locally? If not, I may try to create one, but I have to learn more about Docker first.

@FourMoBro

Couple months ago nvidia pytorch container was almost exclusive one which had compiled working pytorch on CUDA9, with new pytorch release it got easier.

Check out these Dockerfiles: https://github.com/tensoralex/fastai-docker

I appreciate the reply, I really do. I had the NGC setup locally and was able to pull and run their image in a container. As I was playing around this morning trying to simulate the paperspace setup to create a dockerfile, my system updated and now docker won’t run at all. I searched the webs for all of a few minutes and couldn’t find a viable solution so I bagged it. I then decided to just follow the rest of the paperspace script. Everything seems to work for fastai running the lesson1 notebook. Maybe after I finish the course I will revisit the docker setup and configuration dream.

1 Like