Description

Hello,

~

This is a workstation for numerical methods in applied mathematics in industry using the AMD 3970X processor on the TRX40 motherboard.

The workload is both GPU and CPU intensive. The full load is eventually executed on cluster with this machine used for prototyping scratch models. Ultimately multiple compute cards will be added to this workstation for local GPU prototyping before shipping model to cluster; something like 2-4 titan Ampere cards or 2-4 3080 ti cards depending on release schedule and performance.

A small size of fast RAM was chosen alongside with 2x Optane DC P4801X drives using F2FS for models, data, cache, with the Optane 905P using F2FS acting as system drive.

~

The build was quite straightforward with no major problems.

O11 dynamic XL case was thoroughly well built with uncomplicated and well designed structures and systems. Support for large amounts of variable hardware including multiple power supplies. Some screws were not correctly threaded and sit diagonally off center from their holes. Case RGB micro-controller easily physically removed and disabled.

S12A fan installation was largely easy using Noctua rubber fan mounts. Included low noise adapters and Y-splitters were used to group fans into 4 fan blocks. Two fan mounts of forty snapped off when being pulled through and certain angles involved awkward movements in order to gain traction.

Seasonic 1000W power supply installed easily and contained a large amount of additional equipment.

Gigabyte TRX40 extreme motherboard provided many features, accessories, and an excellent level of hardware. Installation of the 3970X CPU was straightforward following included instructions with no re-seating needed. The Noctua NH-U14S CPU cooler mounted easily. The m.2 drives were installed below the motherboard thermal cover which contained pre-mounted thermal pads for contact with drives. Motherboard RGB disables permanently through BIOS.

G.Skill RAM worked with advertised XMP profile at 3800 C14. On-board RGB is on by default with no way to permanently disable or easily control inside Linux.

EVGA 2060 GPU slightly misaligned with rear case slot screws and required force to allow screws to tighten. Fan stop functioned well at idle and overall fan noise is low, however under full GPU load audible coil whine was noticeable.

~

Overall performance for entire build is magnificent at wonderfully low noise levels. No hardware issues were encountered and all components worked as intended with first power-on.

Near future hardware changes include previously mentioned 2-4 titan Ampere cards or 2-4 3080 ti cards when released, change to new TRX40 NH-D15 cooler when released, and additional Optane drives as needed. A larger monitor will eventually replace current monitor as more modern 16:9 32' 4k high refresh panels release.

~

Thank you for reading.

Regards,

  • k80
Log in to rate comments or to post a comment.

Comments

  • 1 month ago
  • 6 points

I just saw a plush seagull build a PC and now I am pumped as ever to build myself a new computer!

  • 1 month ago
  • 3 points

Thank you for the kind response and similarly to the other positive comments. It was a fun build and it is delightful to share with other people who are enthusiastic about computing. The current era of personal computers is quite wonderful.

  • 21 days ago
  • 1 point

I saw a seagull in the thumbnail and was like "yep, we got em"

  • 1 month ago
  • 3 points

A very impressive machine. I wish I had a good reason to build such a beast.

  • 1 month ago
  • 3 points

I was so interested in the seagull I forgot about the rig! +1

  • 1 month ago
  • 2 points

Great PC build! i'm wondering which industry do you work in for applied mathematics?

  • 1 month ago
  • 3 points

I work as a quant researcher in finance, currently primarily stochastic partial differential equations and high dimensional statistics.

  • 1 month ago
  • 2 points

Rad. Good fan choice, too bad they make em ugly.

  • 1 month ago
  • 2 points

Super nice build, components, display and write-up. Thank you. I'm wondering if 32 cores is sufficient for your apps or if the new AMD 64 core CPU would be a more cost-effective solution. BTW, how are the apps running compared to what you were using prior?

  • 1 month ago
  • 2 points

If the 64 core CPU was already released and had comparable single core performance as well as overall platform stability and usability it would have been chosen over the 3970X. Depending on how things go it may be eventually swapped when that CPU releases. Performance levels are very good however for full models everything must be run on cluster eventually as they are well beyond the reach of a local workstation. Overall the CPU is excellent and the platform does not have anywhere near the concessions and annoyances I've been conditioned to become used to accepting on a high end workstation and currently I have nothing but the strongest of praise.

  • 1 month ago
  • 2 points

does that Seagull have a name?

It def reminds me of Dave the Seagull from RandomGamingInHD youtube channel

or the Seagulls from Finding Nemo

Mine? Mine? Mine?

  • 1 month ago
  • 2 points

Hey, congrats for your build! I'm about to build something similar for CFD. Can you please share your temperatures under full sustained load?

  • 1 month ago
  • 2 points

https://i.imgur.com/eyvWrKB.png

Here is the output of lm_sensors at idle and then at load, under full io, memory, and CPU load but not full GPU load. acpitz 1 is measured above ambient but the other outputs are absolute temperatures. Ambient temperature I would estimate around 22-24C.

I have the fan profiles very heavily set to silence and it would be easier to lower these temperatures running the fans more loudly. Even at load you can see the system fans are not engaging beyond 50% fan speeds, as they are rated at 1200 rpm max +-10%, due to me keeping them low. At these fan levels the computer is inaudible at idle until a distance of perhaps 30cm to the ear at which point it becomes audible. At full load it is around half the volume as the most recent X1 Thinkpad with its fans at full; if that is a meaningful comparison to you in any way.

The CPU and BIOS are set to more or less default settings for everything except with XMP enabled and silent fan curves, including auto boosts and auto interconnect frequencies. If I understand correctly this CPU architecture will boost itself until 75C at which point it will slow down, meaning that essentially all cooling setups will go to 75C with these settings but will draw varying electrical loads and will have variable computational performance. By manually tuning or using power limiters it is possible to change this behavior but I have not done so.

With regards to performance, I can say that with these fan settings and BIOS settings this CPU benchmarked in the 90th percentile (top 10%) among 3970X CPUs when I ran it on userbenchmark. This is the only benchmark I've run and my schedule is too busy to do much more than this, moreover Windows was only installed very briefly to check hardware status.

I hope this was helpful to you. CFD is a great field and it contains some of my favorite material. Good luck and I hope your new computer works well.

  • 1 month ago
  • 1 point

Damn, those are very nice temperatures given that you're cooling a 32 freaking cores beast at full load with an air cooler.

Given the fact that noise is definitely not an issue for me, I'm very optimistic considering that I plan to use 9 140 mm fans for the case.

Thanks a lot, you were very helpful!

  • 1 month ago
  • 1 point

Great build I love how in the pictures the bird is building the pc other than that it looks good keep up the great work

  • 1 month ago
  • 1 point

Great build, love the pics.

I need this Threadripper in my life.

I can see with the parts you have chosen you have gone for compute power, future growth potential but also stability (I imagine in your line of work maybe many hrs of compute to get results)

  • 1 month ago
  • 1 point

Why did you choose different kind of Optane drives? Spec seems so be very similar but 905P is less than half price of P4801.

  • 1 month ago
  • 1 point

BTW, I have similar workloads and will be upgrading my main computational rig from 9980XE to 64 core Threadripper. GPUs currently two Titan RTX but will be upgrading as well. I can't add more GPUs because the rest of PCI slots are taken by 960GB 905P cards. My rig is watercooled (custom loop). Originally I was running 2990WX with EKWB AIO cooler but it was very noisy. It's interesting that yours is quite on air cooling.

  • 29 days ago
  • 1 point

If I was u I would not use it for maths, I would use it for minesweeper grinding

  • 18 days ago
  • 1 point

How are the temperatures?

  • 6 days ago
  • 1 point

Nice, I use the same case, motherboard and cooler for my new build, just with the 3960x. Have you checked the thermal test from Gamers Nexus on the Lian Li Case? Best thermals have been achieved with 3 sides and 1 rear fan.

  • 5 days ago
  • 1 point

I'm considering getting the same RAM as you, but the 1.5v draw makes me uneasy, and was wondering if you've used any custom profiles to be able to work around that without losing any performance, or just gone the traditional "set and forget" route of XMP?

  • 4 days ago
  • 1 point

What was your trick fitting in the Aorus Extreme and the
Noctua NH-U14S TR4-SP3 82.52 CFM. Did you change the fan to a 120mm version or did you move the fan more inside?

I guess you used the Noctua NF-S12A PWM as fan cased on the side, top, bottom and rear?

Have you tried disabling the bottom and top ones? Gamer Nexus made a test with the O11 Dynamic XL and this was the best cooling setup. Would be interested if this is also the case in a pretty similar build than mine :)

  • 1 month ago
  • -1 points

This build is great, but $7000!? Im mostly new to the pc building thing, but this is a little over the top and could most likely be built for cheaper. But if you can afford the massive dollar sign, holy hell. This thing is a monster.

[comment deleted]
[comment deleted by staff]
  • 1 month ago
  • 2 points

If you are interested in applied mathematics I would recommend the excellent series "Foundations of Applied Mathematics" by Humphreys, Jarvis. Volume 1 covering introductory mathematical analysis is among the best treatments of the topic and volume 2 will be published early 2020 covering algorithms, approximation, and optimization.