2017 Build, OS and Packages

Part 1 (2017)
1

Let me preface with - I am new to machine learning and deep learning in general…

Jumping in the deep end very early. Current build:
i7 6850k (its the lowest chip that allows 40pcie lanes)
asus rampage v edition 10 mobo (want access to 128 gig ram if needed - m.2 etc)
32 gigs corsair ram ddr4 3200
1T m.2 for boot and hard drive
1T ssd for second OS and storage
1080 ti overclocked asus rog gpu

now the hard part - I will have windows 10 on one drive and ubuntu on the other.

i have been using python via pycharm ide

  1. should i make windows 10 or ubunto my machine learning environment? My guess is ubuntu has the least amount of issues and conflicts with libraries etc…

  2. what do i need to install to have the gpu locked and loaded for machine learning and AI libraries (modules)

  3. should i still use pycharm or run all modules through anaconda? sometimes just trying to install a module with all the dam failures is so frustrating ya want to quit. seems something is always missing.

  4. does anyone know of a reference for getting a clean ubuntu install locked and loaded to the brim with all important machine learning and AI libraries/modules installed…a step by step?

thank you all in advance…

G

2

  1. Ubuntu will be a bit easier as most people test these frameworks out with linux first. That said I am able to run all the code in windows, with very little changes.

  2. Check out the theano or tensorflow docs for that http://deeplearning.net/software/theano/install_ubuntu.html
    For a win installation I had to install the nvidia cuda sdk + and copy and paste in the cudaNN dlls

  3. I’d start with anaconda first, you can always run pip on top of it if the conda command fails

  4. Nope sorry :slight_smile:

3

I really enjoy using manjaro for machine learning (and everything else) since you can easily use all of the arch documentation, yet it is easy to use.