search

VISIONAI / clouddream

Deepdreaming in the clouds: A Dockerized deepdream Guide
370 stars 62 forks source link

GPU support? #15

Open jakelodwick opened 9 years ago

jakelodwick commented 9 years ago

Any interest in adding it? Could open up some very cool realtime (or near-realtime) options.

Here's a preview of the video I'm working on. Each frame takes a few minutes to render on my DigitalOcean Droplet with 2GB Ram.

fluid3b

quantombone commented 9 years ago

Pretty sweet!

There is a GPU-enabled dockerfile for Caffe here: https://registry.hub.docker.com/u/tleyden5iwx/caffe-gpu-master/

And the tutorial on how to get up and running with GPU instances on AWS is here: http://tleyden.github.io/blog/2014/10/25/running-caffe-on-aws-gpu-instance-via-docker/

Unfortunately the cheap cloud providers like DigitalOcean and Vultr do not offer GPU instances, but their instances run you 10-20$/month. The AWS g2.2xlarge instance costs $0.65 per hour or ~500$/month.

Alternatively, if you have to use a CPU, then you can easily spawn ~50 worker nodes on DigitalOcean, have each one process a subset of the frames, then close down the cloud instances.