Description

I decided to put this build together when my 12-core server just was not fast enough to crunch through the machine learning training sets that I was throwing at it. Knowing that big Nvidia GPUs were going to be much more effective for this kind of workload, I put the server up on craigslist and started looking for components.

The Silverstone case and power supply were a natural combo given that the case only accepts SFX power supplies. The ITX form factor, minimalist size, and ability to fit a full-size graphics card sold me on the case. I went for the 500W capacity power supply so as to be able to easily meet the power demands of the system.

The GA-Z170N Gaming 5 motherboard seems to be one of the best mini ITX motherboards available for the Skylake platform. I really liked the quality heatsinks on the components and the availability of thunderbolt ports.

I picked the GTX 980TI video card as it seems to be the best value at the moment. This GPU has almost as many compute elements as a Titan X and performs on par with the Titan X for games. With the release of the new 10 series graphics cards, it was not difficult to find a 980TI on sale.

The processor and memory were all selected for low idle power. The system will eventually be deployed as a server in a compute cluster and the workloads run on it will be bottlenecked by the GPU. The main constraint is then that data must be moved efficiently to GPU memory. Therefore I added just enough memory and selected a low power processor that has support for the latest serial interfaces.

For storage, I did not need a vast amount as most of my data is resident on a NAS box. The Samsung 950 Pro also has some of the best access speeds and good longevity with V-NAND. Not to be overlooked is also how the SSD plugs into the M.2 slot in the back of the motherboard and does not require additional power cables to be routed from the PSU. With the inside of the small ITX box already pretty cramped with the motherboard and power supply, not having to rout SATA power and data cables really helps to reduce the clutter.

Part Reviews

Comments

  • 40 months ago
  • 3 points

I guess you could say it's an...M-IT XBOX! Ha! Haha... ha. +1

  • 40 months ago
  • 4 points

Get outta here with that weak stuff

[comment deleted]
  • 40 months ago
  • 1 point

Hello,

Cool build I was thinking of putting my EVGA gtx 980 ti FTW (slightly more factory overclocked version of yours) in this case, how are you temps? Trying to decide whether it would be a good idea or not.

  • 40 months ago
  • 1 point

I've only monitored load temps under some synthetic benchmarks and I saw numbers in the mid-70s. I also had the case standing up with the GPU oriented on the top. From what I can tell the FTW has the same cooler and runs about 5 degrees hotter, so you will probably be looking at temps in the low to mid-80s.

  • 40 months ago
  • 1 point

well the 950 Pro does have a huge temprature problem so it dos thermal throttle and you did put it in an unventilated part of your case... i wonder if that is a good decesion?

  • 40 months ago
  • 1 point

I am curious about this throttling issue as well. At the moment I am using the case in an upright position in a well-ventilated area, but I will be monitoring the temperatures when I start to do more disk-intensive stuff.

  • 40 months ago
  • 1 point

Results?

  • 40 months ago
  • 1 point

Sick build! Just wondering, what machine learning library are you using? I've only used Tensorflow before, and to my knowledge it only supports Titans and Quadros, does what you're using officially support non-Titan Geforces?

  • 40 months ago
  • 1 point

I am using Theano at the moment. It's very similar to Tensorflow, but Tensorflow has efficiency issues with excess copying on matrix operations that causes it to run 4x slower.

With regards to GPU support, Tim Dettmers has a pretty good write up on a comparison of Nvidia GPUs here: http://timdettmers.com/2014/08/14/which-gpu-for-deep-learning/. It helped guide me to the 980TI.

  • 40 months ago
  • 1 point

Ah I see, thanks!

  • 40 months ago
  • 1 point
I love these Ts CPUs. <3

+1

[comment deleted]
[comment deleted by staff]
[comment deleted by staff]