Build a PC for Deep Learning


Deep learning is hot, it is all over the news, hacker news, that is.

After some research I am convinced deep learning was not just some fad, it is the real deal, I decided to build a new PC for deep learning. (But really, just an excuse I gave myself for replacing my perfectly fine 3 year old PC.)

Here is the part list:

name brand/model/desc # CNY ¥ USD $
case Corsair 780T 1 1279 176
PSU EVGA SuperNOVA 1600 T2 1600w Titanium 1 2899 387
motherboard Gigabyte X99-UD4 1 2699 260
CPU Intel Core i7-5930k 1 4399 570
RAM 1 Kingston HyperX Fury DDR4 2400 32G (16Gx2) kit 3 5097 672
RAM 2 Kingston HyperX Fury DDR4 2400 16G (8Gx2) kit 1 899 122
video card Gigabyte GTX1080 G1 Gaming 3 14997 1770
Hard Disk 1 Samsung 850 EVO 500GB SSD 1 1299 170
Hard Disk 2 WD Black 4TB 1 1999 224
DVD Reader ASUS BW-16D1HT Bluy-ray burner 1 669 84
CPU cooler Phanteks TC14PE 1 399 75
WIFI/Bluetooth Card MSI Wi-Fi AC Kit-S 1 299
Fan 1 Phanteks F140XP 2 168 32
Fan 2 CoolerMaster FP120 PWM 1 79 18
OS Windows 10 Pro 1 986 140
software 1 MS Office 2016 Home-Student 1 499 140
total ¥38666 $4840

The reason I choose this combination of CPU/motherboard is the potential to have 4 graphic cards installed on the motherboard. Core i7-5930K has 40 PCIe lanes that supports 4 GPUs. Gigabyte X99-UD4 has 4 evenly spaced PCIe slots. The PSU I got-EVGA 1600w should be more than enough to support 4 GPUs. Is is unlikely I will need 4 graphic cards, but the potential is there.

I got all parts in China and over-paid roughly 15% compared to US, which is funny because almost all of these were assembled in China.

Here are some pictures (click for full photos):

case (or just box)

Motherboard, PSU, CPU, graphic cards, memories:

Rest of parts, CPU cooler, blue-ray burner, hard-disk, SSD, M.2 Wifi card, fans:

fan, software:

In its full glory: No, I did not do a good job of cable management as you can see above.

This thing is huuuge, here shows the size comparison with a regular mid tower.