WhisperPower - ROG for Computational Fluid Dynamics

[Zka17]

Hello All! How you've been?*:cool:

I've been busy with my work, not too much time to enjoy old habits... However, something came up at work with one my projects: I need a small, compact workstation; easy to carry around, and quiet as possible. I need to run Computational Fluid Dynamics simulations on it - it needs to be multicore, able to run high frequencies, have graphics power, and needs RAM.*

I still have BlackBeauty (https://rog.asus.com/forum/showthread.php?93554-BlackBeauty-reloaded#post652823) - sorry for not being able to complete the picture uploads! But here are the guts: https://rog.asus.com/forum/showthread.php?91762-My-New-System#post638563. The system is till working well, although it has some limitations... it's weight for example... and those fans are keeping it cool, but they are noisy...

So, here is whit what I came up with:
Mobo: Asus ROG Strix Z390-I Gaming
CPU: Intel Core i9-9900k
CPU cooler: be quiet! Dark Rock Pro 4
RAM: G.Skill Ripjaws V Series 32GB 3200MHz CL14
SSD1: Samsung 970 EVO Plus 1TB
SSD2: Samsung 860 EVO 2TB
Graphics: Nvidia GeForce RTX 2080 Ti, FE
PSU: Seasonic Prime Titanium 850W
OS: Windows 10 Pro *

Yes, it is a mITX board! I've had one before (Impact) also for work and it was OK-ish... My initial plan was a Maximus XI Gene (mATX), but had a bad luck with it:*https://rog.asus.com/forum/showthread.php?112547-Need-help-Maximus-XI-Gene&p=780206#post780206. I needed this system up and running, and there is no other mATX board in the ROG lineup, so mITX is for now...

I have finished the build, installed the OS and all of it's updates - then I had to try it!*;)

Started with RealBench... and with pushing the clocks... being careful with the voltages...



This not bad at all! 5GHz with 0 AVX offset at 1.25V, on air cooling - 73C would be an acceptable long term temperature...

And it is pretty stable too:



Well, I know, 15 minutes on RB is not the end of the world - but my previous experience was that this is a good indication of stability... Certainly will do some more testing with other benchmarks - and finally with CFD simulations!

I know... you are waiting for pics of the system... sorry, I don't have them yet... 😞
But stay tuned! :cool:*

[Zka17]

It took some time, but I think it's done... :cool:

I was busy with fixing my Maximus XI Gene board (https://rog.asus.com/forum/showthread.php?112547-Need-help-Maximus-XI-Gene#post780206), then I was waiting for an another CPU cooler...

Here is the final list:
Mobo: Asus ROG Maximus XI Gene
CPU: Intel Core i9-9900k
CPU cooler: Noctua NH-D15S (dual fans)
RAM: G.Skill Ripjaws V 32GB 3200MHz CL14
SSD1: Samsung 970 EVO Plus 1TB
SSD2: Samsung 970 EVO Plus 250GB
SSD3: Samsung 960 EVO 2TB
Graphics: Nvidia GeForce RTX 2080Ti FE
PSU: Seasonic Prime Titanium 850W
OS: Windows 10 Pro
Case: Thermaltake Core V21

Yes, I have changed the Dark Rock Pro4, because the heat pipes were pushing on the memory module closest to the CPU. Not sure why I went with the Noctua D15S though... maybe because of the better compatibility than the D15 (asymmetrical design) - but here it is on the board:











I realize that the D15S suppose to be installed inverse, with the bigger part toward the top of the board, but on this board the PCIE x16 slot is lower than usual (there is PCIE x4 first) - so it fits tis was too! I've got the Chromax things too, but I am not 100% happy with them... Noctua skipped a little bit on design here: the sides of the covers does not go down all the way (they did not think of horizontal setups, on vertical placement of the board this won't show).

Then I installed everything in the case... I choose the Thermaltake Core V21 because it allows the horizontal placement of the board (I prefer this setup with such big coolers), and because it is pretty small/portable. Here is how the stuff looks inside:





Yes, the cable management is basic: it is out of the way of the airflow - it is good enough for me! Also upgraded the D15S with black fans - two NF-A15 HS-PWMs, then added a third on the back of the case. Unfortunately they are loud! Compared to the be quiet Silent Wings 3 ones on the Dark Rock Pro 4... but at least they have way better airflow - will see the temps later!

I did use two 140mm Silent Wings 3s on the top though - to get rid of the hot air produced by the graphics card:



And here is the closed case:









Well, the windowed panel did not came out right - will need to figure out how to picture reflective surfaces... but, on the side of the graphics card is a vented panel with filter, and on the top a vented panel without filter.

That's it for now... let me know what do you think!

Will be back with some benchmarking and temps...*

[R5Eandme]

Thanks for the great pictures of this interesting and meticulous build. How did you manage to fit that monstrous NH-D15S dual fan cooler onto an mITX board in that small case? 😄

I imagine that the computational fluid dynamics simulations that you are running involve a great deal of floating point calculations, maybe dual precision floating point? I used to do a lot of statistical analyses and so I am very interested in what are good CPUs for efficient floating point operations (FLOPS per cycle is one measure of efficiency).

Your i9-9900K is a Skylake-Coffee-Lake cpu capable of 16 FLOPS/cycle dual precision and 32 FLOPS/cycle single precision which is not bad. The Skylake-X series are probably the top consumer-level cpus for floating point efficiency (32 FLOPS/cycle dual and 64 FLOPS/cycle single precision) but are much more expensive.

See this article on how the different Intel and AMD cpus compare with respect to floating point math:

https://en.wikipedia.org/wiki/FLOPS

Of interest is how much better the top Intel cpus are at floating point than are AMD. For example, Ryzen Zen+ (e.g. 2700x) can do 8 FLOPS/cycle dual and 16 FLOPS/cycle single precision. That is half the efficiency of your Coffee Lake. However, the new Zen 2 (3000) series can achieve 16 FLOPS/cycle dual and 32 FLOPS/cycle single precision, comparable to the Coffee Lake.

So I am wondering if there is a trend at AMD to improve their CPU performance with regard to floating point computation? Can we see in the future floating point performance comparable to Intel but at better prices? Anyone with knowledge of this, I would appreciate your comments as this kind of info seems hard to come by as floating point efficiency is not a priority for gaming as much as it is for niche scientific work stations.

Understand that I am not trying to hijack the thread, I just thought it relevant for your work and that you would be interested in this line of discussion. If not, I will respectfully move it to a new thread.

Thank you.

[Zka17]

Thanks! Fyi - the Maximus XI Gene is an mATX board... the Core V21 case is pretty spacious at these form factor level... and the D15S fit almost everything!

Regarding the FLOPS, you caught me there! I do not *understand too much of it, and have no idea if the softwares I am intending to use on this system are using it or not. Maybe you could check it for me: SimVascular and Ansys Fluent. I am just a user, with a little bit of understanding of computers...

I have double-checked the system requirements for the above applications, and specifically the Fluent benefits of CPUs with up to 16 cores, high speed CPU and RAM, wide bandwidth and high amounts of RAM and can use the GPU's computational power too.

I am aware that this system, I built, is not really a workstation - it is only a temporary solution. It is rather an experiment... targeting the small form factor (for portability), quietness, and speed. I do expect/waiting for the new Threadripper CPUs though... more affordable than Intel, high core count, and similar per core computing capacities with the new Ryzen... will see!

Do you know a benchmark for the FLOPS? I would be interested to run some tests...

And please, keep commenting on this subject - I am interested!*

[R5Eandme]

Hello Zka17, thank you for the extra descriptions of your mobo and case. You build looks like a fun and powerful machine.

I wish I was expert in floating point operations as it relates to software and cpus, otherwise I could help you more! That is why I posed the question in the first place about which CPU is better for floating point operations, Intel vs AMD.

However, I did visit the websites for Sim Vascular and Ansys Fluent software. I learned enough to see that my question was a bit naive. Allthough I strongly suspect that these models are solved using floating point arithmetic, there is much more to running these simulations efficiently than using a processor that is fast with floating point math. It looks like at least Ansys takes advantage of parallel computations such as can be run on the thousands of graphical processor units in the video card. It looks like they model fluid dynamics as a collection of "cells" or units that coursely approximate the behavior of real fluids like blood in the heart and vessels. So GPU processing cores must be at least as important as floating point speed for applications that can take advantage of GPU parallel processing. Maybe the CPU isn't as critical as the GPU? You might ask them if the software would run more efficiently on a workstation graphics card such as a Quadro or FirePro compared with a gaming/general purpose graphics card. Or maybe your new RTX 2080Ti is a match for those workstation video cards at running these CFD models? I believe that workstation video cards are faster at floating point math, and are used for Computer Assisted Design (CAD) software. I found an interesting link at AMD on serial vs parallel processing for High Performance Computing

https://www.amd.com/en/technologies/hpc-explained

[Zka17]

Some temperature comparison of the Noctua NH-D15S on the Maximus XI Gene with the Intel Core i9-9900k at 5GHz, 1.25v:

*

Initially I setup the system on the box of the motherboard (open air) with the original fan on the cooler. Let the AIDA64 run for 5 minutes in idle, then run the stress test for 5 minutes, then wait 5 minutes and save the results.

The original, 1 fan setup of the D15S outside the case produces pretty decent cooling... average of the max temps is right below 80C (79.125). Adding a second fan to the cooler, but still outside the case lowers the average of the max temps to about 73C (72.875) - that is an improvement of almost 10%! However, placing the whole thing in the Core V21 case*(with the dual fan setup, but adding a 3rd exhaust fan in the back), does increase the average max core temps to about 74C (74.25)... I think, I can live with that!

Interestingly, the dual fan setup on the cooler does have a beneficial effect on the average motherboard and VRM temperatures too: 7.4C and 9.9C lower, respectively. Inside the case these temps will slightly increase too, but again average mobo temp at 25C and VRM temp at 34.4C are still pretty good!*:cool:

[AS185]

The Noctua cooler is one of those components that is competing on the market with water cooling kits. I didn't know it was that big. Nice Zka!

[kkn]

hehe, huge bugger these coolers