Description

Dual boot setup: Windows 10 and Ubuntu 16.04.

Built this to develop my skills in deep learning. Using Tensorflow and building some CNN and RNN models for work. Getting better at utilizing 12 core / 24 thread cpu in development. Threadripper! I'm overclocking to 4.0 GHz and so far so good.

Playing Overwatch and CSGO! Getting destroyed in Fortnite. Hearthstone casual. I think gaming with a 144Hz monitor is awesome! Once you start playing at the higher refresh rate, you can't go back. I'm not sure if I would go with this CPU if my only purpose was to game. I swear I see slight hiccups when gaming and I think it has something to do with with the way AMD and the memory work. I've messed around with Ryzen Master a little bit.

Hard disks overboard. I started just with the the Samsung EVO 960 for Boot/OS. I recently added the Western Digital Black NVMe for more "faster" space for data. I was pleasantly surprised by the WD. Current benchmarks show the WD is faster than the Samsung. The other drives are now just for more data storage.

RAM is good so far. I actually had to RMA the first kit I bought. I had system instability for awhile and it took forever to figure out what was wrong. Eventually, I booted into a memtest86 usb drive and found errors in the RAM. This was my biggest headache for this build.

Overall, I'm super happy about this build! Of course, when you get to these prices on a build, you're going to end up with an awesome machine.

Part Reviews

CPU

Right now I'm overclocking to 4.0 GHz. So far the system has been stable.

Storage

Initial benchmarks have been amazing. This is a really fast drive. Faster than my Samsung EVO 960.

Edit: I knocked this drive down a star. I've struggled to get this drive working in Ubuntu 18.04 (Freezes system when trying to access). After searching the Western Digital and Ubuntu forums, it doesn't look like this drive works well with Ubuntu right now. Not sure if it's drivers or bios or something else. If anyone has luck getting this to run in Ubuntu 18.04, let me know. Still works great on Windows 10.

Monitor

I LOVE LOVE LOVE this monitor. 144Hz gaming all the way!

Comments

  • 17 months ago
  • 1 point

Cool rig, same monitor as mine, totally right about high refresh rate: once you go full 144 fps you litterally cant go back. What temperature do you have on the threadripper with the h100i cooler? Im using it for my 8700k@4.9GHz and it seems a bit to high temperature, so i went back @4.7GHz.

  • 17 months ago
  • 1 point

What was your temp at 4.9GHz?

  • 17 months ago
  • 1 point

@4.9GHz i had 45/49C when idle. i stopped my first bench when i saw a 90+C on 2 cores and the cooler fan where 100%. Plus while gaming it was pretty unstable after 10/15 minutes i always got crash or bluescreens.

  • 17 months ago
  • 1 point

Then you get G-sync and you can't ever have another monitor without it!

  • 17 months ago
  • 1 point

4.9 GHz is crazy! Even 4.7 GHz is crazy. I'm jealous. To answer your question, when I ran prime95, I ran it for about an hour and the peak temperature on the cpu got to 74 C. I'm not a prime95 guru, but different tests caused different temps. It would be stable at 66 C, then a different test it would go to 74 C, then another test would drop it to 64 C. A little uncomfortable for me, but I think it never got too dangerous for the cpu. CPU is at 1.35V. Also, I could see the GPU fan go from 0 to 240 rpms, but I wasn't even using it, so you could tell inside the case the temperature was getting warmer. You definitely can feel the heat coming off the radiator/fans. The H100i fans bumped up to 2400 rpm and the pump was running to 2800-2900 rpms.

Overall, so far so good. Passed the prime95 test for an hour. I know a true test for stability is way more than that (24 hours+), but I haven't had the heart to let it go for that long.

  • 17 months ago
  • 2 points

4.9GHz full-core is crazy but with a bit of luck and overvolt you can reach it on a 8700k. 4.7GHz full-core is a pretty standard overclock, remember the stock i7-8700k has a single core 4.7GHz overboost, 4.5GHz o 2 cores and so on. Dont be jealous, u have 12cores/24 thread @4GHz it way better than my 6cores/12threads @4.7 :D Pretty cool temperature too as far as i know.

  • 16 months ago
  • 1 point

How has the cooler performed with low surface coverage on the TR4? I'm dubious about AIO cooling solutions but the Noctua is an efficient, nasty looking beast.

  • 16 months ago
  • 1 point

I'm considering a similar machine for Computer Vision and Deep Learning. Did you have any problem dual booting on Windows 10 and Linux? Which Linux distribution did you use? Was the intergrated WiFi detected correctly? Any drivers problems you would mention? Did you have any problems using computation libraries on AMD instead of Intel? Thanks in advance for the feedback you will want to share!

  • 16 months ago
  • 1 point

Hi,

I would buy different RAM. I've been having trouble getting the full speed with my motherboard/cpu combo. I've been talking with MSI about it. So annoying.

I also had issues with the Western Digital NVMe drive in Linux. I ended up exchanging it with another Samsung EVO 960 because that worked right away in Linux. I just recently upgraded to Ubuntu 18.04. Everything went pretty smooth with that. Getting CUDA setup correctly was a little tricky in Ubuntu but once you get the right Nvidia drivers running, it worked for me. I'm just using Tensorflow (GPU) and can't tell you about any other library (I have thought about checking out PyTorch though. It's been gaining a lot of popularity).

I haven't really ran into anything else too troubling. I've been running Ubuntu more than Windows. I did setup Tensorflow (GPU) on Windows and didn't have any issues with that but mainly develop in Ubuntu. I game on Windows and that's running pretty smoothly now too. I was seeing a little micro-stuttering in CSGO but I locked Steam/CSGO to a set of CPUs and that seemed to fix it. I think the micro-stuttering was happening when the game threads bounce between the AMD cores (not 100% sure).

Overall, I'm happy and it's been very stable for me (prime95 tested)

Good luck and be sure to post your build! Link me when you do it!

  • 16 months ago
  • 1 point

Thanks for the report, chillyice9! I'm going to use the same RAM on the ASRock X399 Taichi. The RAM is in that board QVL. I attempted to find the QVL of your motherboard, but I wasn't able to. I'm going to use two smaller SSD from Samsung, so I should be ok, then. Do you have any experience with virtualization and "PCIe (GPU) passthrough" and Deep Learning? I was planning to try run both Windows and Linux on top of unRAID OS (which is a thin Linux OS dedicated to NAS and virtualization), or just using Linux as host and Windows as guest. unRAID OS is used in multi-user gaming machines with less than 5% worst case penalty, but the typical usage of the GPUs in gaming and Deep learning are different, and I'm worried that the latter kind of applications could incur in a higher penalty. I would like to eliminate the need of rebooting, if possible. I'll let you know when I build the machine. Thanks again!

  • 15 months ago
  • 1 point

Sorry for the slow responses, I haven't check this site in a bit. Sounds good. I don't have experience with what I think you're attempting to do. I have read about some solutions to have different os/virtualizations and running the gpus. I think VMWare has something you can get to the 'bare' metal. I haven't hear of unRAID OS but I'll read about it. Definitely let me know when you get your build done! Sounds awesome!

  • 15 months ago
  • 1 point

PCIe passthrough reportedly had problems on Threadripper in the recent past. I'm not sure whether they're fully resolved; you may have to upgrade to a newer kernel.

  • 9 months ago
  • 1 point

Well for deep learning I would say your cpu is great, over clocking it or upgrading it won't do much for you where deep nets are concerned, I would upgrade the gpu and have 2-4 of them. If u use pinned memory then your mini batches pass directly to the gpu and if not faster ram improves 0-3%, it's better to have more ram at least equal to gpus. Can't use 16bit models on gtx only 32bit, with RTX cards you can run 16bit models that are twice as big with the same memory compared to gtx .

  • 9 months ago
  • 1 point

Thanks for the info. Yeah, I've been eyeing the RTX cards. I might pull the trigger at some point this year.

  • 9 months ago
  • 1 point

That's pretty much the only good use for those expensive cards. The performance is similar to the last generation, but the prices are 2x larger. For deep learning, however, they are beautiful machines :)

  • 12 months ago
  • 0 points

it says he overclocked it, but in reality that 4.0ghz frequency is actually normal for this cpu

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