Wanted to start building my own deep learning machine for IOT projects. Been using it for the last 6 months. Took me a two days to put these together and have my first convnet up and running. The user manuals and youtube were great help to build the server from scratch. The Corsair liquid cooler has been holding up pretty well for large load operations. Apart from the setup i have WD 4TB drive to other project or large data files. Active Applications and Active projects are stored on SSD.
What worked. 1. The liquid cooler 2. Fully Modular ATX: Though 1200W seems extensive, the setup requires 750W, but wanted to have more power in case of an additional GPU. The flexibility the fully modular PSU provide is exceptional. What can be improved. 1. MSI Motherboard: Overall the motherboard was good, but it kept generating different error codes although the components were working. Spoke with the Technician and they confirmed it was OK and the error codes would "settle in" after having it run for a couple of days. 2. Definitely look for deals while purchasing.
One thing i would change would be to add another GPU to speed things a bit. Based on the amount of time i spend on this server, TCO is wayyyyyyyyy cheaper than Amazon or Floyd Hub type subscriptions.
Here is a link to Tensorflow Install with Win10 GEForce 1080GPU. https://github.com/kartikvega/TensorFlow-Install-on-Win-10-w-1080TI