Description

First watercooling build (in progress). I'll be using this for machine learning/deep learning and occasional gaming.

Log in to rate comments or to post a comment.

Comments

  • 22 months ago
  • 3 points

Cool build! At least your not using all those graphics cards for mining. +1

  • 22 months ago
  • 3 points

That moment when you ram is twice as expensive as your cpu... Great part choice though, can't wait to see it finished.

  • 19 months ago
  • 1 point

I mean, the ram is more than twice as expensive as an i7-8700k, a pretty beastly CPU. So that's the scale of this beastly machine.

  • 22 months ago
  • 2 points

Wow what an amazing pc

  • 17 months ago
  • 2 points

Hol'up, what is that application of thermal paste. Besides that amazing build.

  • 22 months ago
  • 1 point

MORE PICS!

  • 22 months ago
  • 2 points

Added pics of my progress so far. Successful POST! Next up, adding GPUs and storage, then onward to the water loop.

  • 22 months ago
  • 1 point

Nice! Now were talking. Looks good. Step by step photos are the best. +1

  • 22 months ago
  • 1 point

Just a quick progress update with a couple of photos. GPUs, radiators, and fans are all installed. I ran into a small hitch with the Thermaltake Riing Plus 14 fan pack and Thermaltake Commander RGB fan controller. Turns out the RGB fans use USB headers rather than standard 4-pin fan plugs, which in retrospect makes sense and is great for software control of the LEDs, but means that the fan controller won't be able to control fan speed for the case fans. I've set it up so that the first channel of the fan controller controls the fan speed for the CPU loop radiator while monitoring the temperature of the CPU loop coolant, while the second channel controls the GPU loop fan and temp. I've attached the thermal probe for channel 3 to the CPU heatsink on the back of the mobo, and channels 4-6 are monitoring the GPU heatsink on the back of the GPUs. So I was able to use the temperature functions of the other 4 channels, but not the fan speed controls. I might revisit this down the road to see if there's a software or hardware solution to control the RGB fans from the controller, but for now a software-only solution is fine. I made decent progress on the water loops, which I've planned out fully but I'm waiting for more tubing to complete. I've also ordered more 90° rotary loops in chrome, some chrome spacers for the GPU loop, and some female-to-female extenders to place in the rubber grommets to clean up the connections between the front and back of the case.

  • 21 months ago
  • 1 point

I'm a little confused... Why would you choose a TR 1900x instead of a Ryzen 7 2700x?

2700x - Better turbo frequency - better lithography, lower power consumption - has a cooler - like $100 cheaper -

TR 1900x - better base frequency (by .1) - can support 1TB of memory (you used 64 GB)

In benchmarks, the TR 1900x gets outperformed by the 2700x in all categories

http://cpu.userbenchmark.com/Compare/AMD-Ryzen-TR-1900X-vs-AMD-Ryzen-7-2700X/m340638vs3958

  • 21 months ago
  • 3 points

Multi-thread performance is very important for deep learning. The TR1900 matches or beats Ryzen 7. http://hwbench.com/cpus/amd-ryzen-7-2700x-vs-amd-ryzen-threadripper-1900x Plus, by going with the TR, he leaves an upgrade path to the TR 1950, if he wants to double his CPU power in the future.

  • 21 months ago
  • 1 point

The multi-thread performance is within the margin of error, so I don't see much speculation in that. Although I do agree with the upgrade path, the Ryzen 7 2700x is the newest of its series and as of now has no newer model, while the TR 1900x has the 1920x and 1950x. Thanks for the clarification

[comment deleted]
  • 20 months ago
  • 1 point

sick sick

  • 16 months ago
  • 1 point

Personally I would have gone with EK...

[comment deleted by staff]
  • 22 months ago
  • 3 points

I don't think so, but I'm no confident. I think the Titan V's significantly better than the 1080ti; but it's teslor cores are hard to take advantage of (even outside of gaming), and 3 1080ti's probably have it beat for deep learning.

Edit: Quick google search says about what I thought; but it's still not a reliable answer. https://medium.com/@u39kun/titan-v-vs-1080-ti-head-to-head-battle-of-the-best-desktop-gpus-on-cnns-d55a19866b7c

Time will tell how much improvements we will see at the framework/CUDA/CuDNN level to exploit the Volta GPU capabilities as much as they can, but the initial numbers that I have observed on popular CNN’s... don’t seem to justify getting a Titan V...

  • 22 months ago
  • 3 points

Great. I edited in a quote and source, but didn't fix "I'm no confident".

Now i'm gonna have confident issues all day.

  • 22 months ago
  • 1 point

Nah, don't have confident issues. It's a funny but simple mistake.

  • 22 months ago
  • 1 point

I'm not worried about it. I'm pretty good at keeping things under wraps.

It sure is good thing I don't have confidant issues.

<insert Seinfeld theme>
  • 22 months ago
  • 2 points

So, great observation, and I gave this some thought as well (I put this comparison together while I was researching: https://imgur.com/wEwyeZ1). The Titan V does have ~35% more bandwidth than the 1080 Ti at base clocks, but at ~3x higher cost. I'm usually running multiple model configurations concurrently, so with my budget it seemed to make more sense to go with 3 1080 Ti's rather than one or a pair of Titan V's. I also like that the Aorus cards come with a factory waterblock and 4-year warranty, and they should hold resale value fairly well for a year or two if I choose to upgrade.

  • 19 months ago
  • 1 point

The Tensor cores in Titan V are very good for RNNs.

[comment deleted by staff]
  • 21 months ago
  • 3 points

Machine learning software does not yet take advantage of the Titan V capabilities. It will be several more months. Right now, the 1080ti gives almost equal performance at 1/3 the price. That may change within 12 months, but you can always add a second card if it becomes worth it.