Assemble $3000 deep learning machine with 1080 Ti, i7-8700, Z370 and 64GB memory

Specs

PCPartPicker part list / Price breakdown by merchant

Type Item Price
CPU Intel - Core i7-8700 3.2GHz 6-Core Processor $319.99 @ B&H
CPU Cooler Corsair - H150i PRO 47.3 CFM Liquid CPU Cooler $159.99 @ Amazon
Motherboard Asus - ROG STRIX Z370-E GAMING ATX LGA1151 Motherboard $197.99 @ Amazon
Memory Corsair - Dominator Platinum 64GB (4 x 16GB) DDR4-3200 Memory $799.99 @ Amazon
Storage Samsung - 970 Pro 512GB M.2-2280 Solid State Drive $229.00 @ Amazon
Storage Samsung - 970 Pro 512GB M.2-2280 Solid State Drive $229.00 @ Amazon
Video Card EVGA - GeForce GTX 1080 Ti 11GB FTW3 HYBRID GAMING Video Card $749.99 @ B&H
Case Corsair - Air 540 ATX Mid Tower Case $119.99 @ Amazon
Power Supply Corsair - 1000W 80+ Platinum Certified Fully-Modular ATX Power Supply $198.99 @ Amazon
Prices include shipping, taxes, rebates, and discounts
Total $3004.93
Generated by PCPartPicker 2018-08-04 12:12 EDT-0400

PCPartPicker part list / Price breakdown by merchant

CPU

I assembled this rig right before Intel’s 9th-generation processor launched. Obviously, the best option for me is i7-8700k. But since this is my fist assembled rig, I want to keep it simple and keep overclock away. Without overclocking, i7-8700k give nearly the same performance as i7-8700. Therefore, i7-8700 is good enough for me.

CPU Cooler

To be honest, for i7-8700, air cooler should be more than sufficient. But I already suffered enough fan noise from my laptop, so I decide to get this luxury liquid cooler.

It is worthy. So far, it is just as quiet as breath. ( Just for daily use case. I have not tested its limit)

Motherboard

It looks like, for the time I assembled this rig, Z370 is the only option for Intel’s 8th-generation processor. It is said that Asus motherboard have easily configured BIOS. This is an attractive selling point for a newbie.

The only difference between Z370-E and Z370-F is that Z370-E have wifi module. However, why a desktop with wired connection need wifi? As it turn out, I paid the price of Z370-E to get Z370-F functions and left the wifi antenna in box.

Memory

Gain max supported size(64 GB) to settle once and for all. As for the clock, the higher the faster, but also the high price. As a compromise, 3000+ would be good enough.

In most case, 64 GB memory would be a waste of money. But it would pay off when you working with large datasets. In fact, the dataset size would grow rapidly after feature engineering. What is more, packages like Pandas would only perform its functions with memory size that is twice of the dataset size.

Storage

Today, Samsung 970 Pro is one of the best M.2 NVMe SSD, and I get two to dual boot Windows 10 and Ubuntu 18.04 on separated SSD.

As for now, it is quite obvious that the M.2 NVMe SSD would be the most promising SSD.
If you are looking for SSD now, it is probably that you would be confused by these concepts: M.2[1], SATA[2], PCIe[3], NVMe[4].

To simplify (at the cost of precision):[5]
Both SATA and PCIe are bus[6],a communication system including connectors(interfaces, sockets), lanes and protocols(standards).
SATA use memory as medium while PCIe communicate with CPU directly, so PCIe could be much more faster than SATA.[7]

M.2 is just a connector(interface, socket), and it can use SATA lanes or PCIe lanes.
NVMe is just a protocol(standard), and only PCIe lanes can support NVMe.

GPU

It is said that GeForce GTX 1080 Ti is the most cost-effective[8][9] for deep learning and EVGA have worldwide warranty.
I select FTW3 HYBRID because I want it to be quiet (liquid cooler).

Case

Obviously, Z370 motherboard need an ATX Mid Tower case. It is said that large cubic case is good for cooling.
As it turns out, all the computer components are perfectly(compactly) fitted into this case.

PSU

According to pcpartpicker, the estimated Wattage of this computer is 167W - 508W.
For security(Psychological placebo), doubling the estimation should be more than sufficient.

Assemble

Warning: No video, No photo.

As a newbie, it is quite troublesome to assemble a rig. However, at the moment you eventually assembled this, it is said that you would gain a sense of achievement. But sorry, I only felt tired.

  1. Take photo(not only in the fist step)
  2. Unpack
  3. Read and follow the manual of case, motherboard, CPU cooler, PSU (Of course not in the same time but in the order)
  4. CPU
  5. Fix motherboard in case
    • Install CPU cooler backplate before that
  6. CPU cooler
    • Put the fans in front panel since this is the only place compatible with 3 * 120 fans.
    • Put the tubes in bottom, otherwise it would block the upper fans.
    • As for fan holders, pump tach cable connect to AIO_PUMP, two fans connct to CPU_FAN and CPU_OPT, and the last one connect to the cooler connector. (Or you can try that, pump tach cable connect to CPU_FAN and all three fans connect to cooler connectors, but the fans would be slowed down[11])
  7. Memory
  8. SSD
    • According to the manual, the first M.2, which located at right bottom, have better cooling. So later I would install Ubuntu 18.04 on this M.2 and install Windows 10 on the second one.
  9. Fans
    • 2 * 140 fans can be put in upper panel(the case already installed them)
  10. PSU
    • Shunt the cables through different holes
  11. GPU
    • GPU is huge and would block the USB 3.1 connectors and the cables arrangement. So just design the cables arrangement before installing GPU.
    • Put the GPU fan in rear panel since this is bigger than the same size air cooler fan and would not fit in the upper panel.
  12. Congratulation! You get a black box.

Reference

[1] https://en.wikipedia.org/wiki/M.2
[2] https://en.wikipedia.org/wiki/Serial_ATA
[3] https://en.wikipedia.org/wiki/PCI_Express
[4] https://en.wikipedia.org/wiki/NVM_Express
[5] https://www.quora.com/What-is-M-2-PCIE-NVME-SATA-III-SSD
[6] https://en.wikipedia.org/wiki/Bus_(computing)
[7] http://diy.pconline.com.cn/858/8588100.html
[8] http://timdettmers.com/2017/04/09/which-gpu-for-deep-learning/
[9] https://blog.slavv.com/picking-a-gpu-for-deep-learning-3d4795c273b9
[10] https://medium.com/mlreview/choosing-components-for-personal-deep-learning-machine-56bae813e34a
[11] https://zhuanlan.zhihu.com/p/26772358