Dream Machine for Deep Dreaming

[Rafael Espericueta]

I would like to put together my dream machine for experimenting with deep neural networks accelerated via all those lovely CUDA cores. What hardware configuration is recommended for this?  I've had so many hardware incompatibilities in the past; hopefully, starting out with the right system will be a good start.  

Has anyone got this stuff all up and running on an Ubuntu 14.04 system, using Anaconda, with up to date hardware?  What would you recommend?

I'm thinking of an NVIDIA Titan or a GTX 980, as a second GPU.  I thought it would be best to have one GPU for graphics and another for deep neural networks. 

What motherboard do you recommend?  What CPU?   How many CPU cores?  Anything else?!   THANKS!  :-)

[robbyl...@gmail.com]

You can start experimenting with amazon ec2 instances for about $0.2 a hour for a spot gpu instance. 

You can buy your own machine, but if you are just playing around with already trained networks, what ever you wrote might be an over kill just for deep dream. 

I don't think lot's of cores would make a big difference. Everything runs on the GPU anyhow. One good GPU would also probably be enough. 

[Nanne van Noord]

I found this blog post very informative: https://timdettmers.wordpress.com/2015/03/09/deep-learning-hardware-guide/

[Rafael Espericueta]

Thanks!   That is exactly what I was looking for.    :-)

[Rafael Espericueta]

Thanks robbyl, but I want to do more general work with deep neural nets, besides deep learning on trained networks. (I am also trying to set up a virtual machine as you described.)