Thanks for sharing.
My main intention is to start with 1 GPU (GTX 1080 Ti) , but gradually add 3 more, so the box should support 4 GPUs at the same time ( i do not need all GPUs to be connected each other, but mainly for running 4 different experiments on each of them for hyper parameter tuning).
I hope to get full PC build with 1 GTX 1080 Ti 2000–2500$, cheaper better.
I have chosen these parts so far (same as your config suggestions):
Any recommendations for
a) CPU Cooler,
c) PC (not rack) Chassis ?
As I said, my goal is mainly Deep learning NLP, not gaming or image processing, so if you think you can replace some parts(except GPU) with cheaper ones which will still support 4 GPUs in future, please feel free doing so.
I have my own 1 TB SSD.