Ansys Workstation Build

Kurt Clark

n00b
Joined
Mar 14, 2017
Messages
11
I am trying to build a humble workstation for Ansys simulations and also Solidworks CAD.
My build is here

PCPartPicker part list: https://pcpartpicker.com/list/w4KyvV
Price breakdown by merchant: https://pcpartpicker.com/list/w4KyvV/by_merchant/

CPU: Intel - Core i7-8700K 3.7GHz 6-Core Processor ($414.89 @ B&H)
CPU Cooler: Noctua - NH-D15 82.5 CFM CPU Cooler ($89.95 @ Amazon)
Motherboard: Asus - Prime Z370-A ATX LGA1151 Motherboard ($167.99 @ SuperBiiz)
Memory: G.Skill - Trident Z 32GB (2 x 16GB) DDR4-3200 Memory ($459.99 @ Newegg)
Storage: Samsung - 960 EVO 250GB M.2-2280 Solid State Drive ($127.89 @ OutletPC)
Storage: Seagate - Barracuda 2TB 3.5" 7200RPM Internal Hard Drive ($66.88 @ OutletPC)
Video Card: PNY - Quadro P4000 8GB Video Card ($799.99 @ B&H)
Case: Cooler Master - CM 690 III ATX Mid Tower Case
Power Supply: SeaSonic - M12II 650W 80+ Bronze Certified Semi-Modular ATX Power Supply ($167.83 @ Amazon)
Total: $2295.41
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2017-11-29 12:48 EST-0500

I have a few questions regarding this build and hope the inteligent folks at HardForum can help me out with this

  1. Is i7 8700k (6 core) ok for Ansys simulations / CAD or should I go in for Ryzen7 1800x (8 core) and a compatible motherboard. I was more leaning towards the Ryzen7 due to the 2 extra cores but opted out at the last moment.

  2. I have gone with Intel CPU currently with the reasoning that most simulation / CAD software will be more tuned to Intel than AMD Ryzen. Is my thinking flawed?

  3. Is my selection of Motherboard / Ram upto the mark do you notice any bottlenecks?

  4. Is my psu selection 650W, ok for the power or is it too overated. PCpartpicker shows my power consumption as 330W. Would a 550W be fine? I will initially be running the PC at stock speeds and overclock as needed.

  5. I am opting for the Noctua NH-D15 over the Hyper 212X is this way overkill?

  6. Case I am currently opting for Cooler Master CM690III, but its a tad expensive, is there an equivalent / alternative. I mean one where the Noctua would fit in.
TIA
 

Derfnofred

Gawd
Joined
Dec 11, 2009
Messages
599
Given the cost of Ansys, it wouldn't hurt to call them up and get recommendations on what actually serves you best. E.g. I know (and spec'd for work, not the system update thread below) that COMSOL likes memory and bandwidth far more than cores or even core speed (memory's the bottleneck), so I opted for less CPU in my budget to up the memory. Ansys and CFD problems in general may be more amenable to cores. which may make the 1800x do better, but given there are Intel "whitepapers" on Ansys optimization, methinks there's going to be some serious advantages to staying on Intel's pipeline, if purely from a compiler perspective.

My bet is you want fewer faster cores still and as much memory bandwidth as you can get. I'd honestly think about going to a quad channel motherboard and a lesser processor. E.g. I bet an LGA2066 platform would ultimately be faster. Not because of the core/processor speed, but that speedup on the memory side.

* And maybe not LGA2066 since the "quad channel" on that is fraught with implementation issues. Threadrippers are cheap right now, and, even if you're not getting the most out of your cores (plenty of the sims don't scale to processors well), the memory bandwidth on the architecture should be of note.
 
Last edited:

Engr62

Gawd
Joined
Mar 24, 2015
Messages
638
If you are planning on getting the HPC license for ANSYS so you can take advantage of multiple cores, you will probably be better served to get the i5-8600K instead of the i7-8700K and save some money. When using multiple cores, ANSYS runs slower when hyperthreading is enabled. ANSYS saturates the cores, so the overhead associated overlapping the two instruction streams will actually slow the solution time. If you have an i7 or i9, you'll want to disable hyperthreading in BIOS.

If you really want to speed up your solution, you may want to invest in a Nvidia Tesla GPU (~$1500). These usually have no video out, and are merely for processing.
 

Kurt Clark

n00b
Joined
Mar 14, 2017
Messages
11
My bet is you want fewer faster cores still and as much memory bandwidth as you can get. I'd honestly think about going to a quad channel motherboard and a lesser processor. E.g. I bet an LGA2066 platform would ultimately be faster. Not because of the core/processor speed, but that speedup on the memory side.
* And maybe not LGA2066 since the "quad channel" on that is fraught with implementation issues. Threadrippers are cheap right now, and, even if you're not getting the most out of your cores (plenty of the sims don't scale to processors well), the memory bandwidth on the architecture should be of note.
I was looking in the Intel Core I7-7800X 3.5Ghz 6core for that. It has quad channel and supports upto 128GB ram and just a bit more priced than the i7-8700k. I don't know why I finally decided on the i7-8700K, probably since its 8th gen and I thought it would be a tad bit faster that the 7800x.
 

Kurt Clark

n00b
Joined
Mar 14, 2017
Messages
11
If you are planning on getting the HPC license for ANSYS so you can take advantage of multiple cores, you will probably be better served to get the i5-8600K instead of the i7-8700K and save some money. When using multiple cores, ANSYS runs slower when hyperthreading is enabled. ANSYS saturates the cores, so the overhead associated overlapping the two instruction streams will actually slow the solution time. If you have an i7 or i9, you'll want to disable hyperthreading in BIOS.

If you really want to speed up your solution, you may want to invest in a Nvidia Tesla GPU (~$1500). These usually have no video out, and are merely for processing.
This is an eyeopener. I had read somewhere that hyperthreading has to be disabled for Ansys but hadn't looked for the i5-8600k its a good idea and go me thinking. Apart from i7 being 6C/12T vs the i5 6C/6T isn't there any other difference between them?

Regarding the Tesla GPU you mean in addition to the Nvidia Quadro card also putting in a Tesla GPU would improve performance?
 

Engr62

Gawd
Joined
Mar 24, 2015
Messages
638
This is an eyeopener. I had read somewhere that hyperthreading has to be disabled for Ansys but hadn't looked for the i5-8600k its a good idea and go me thinking. Apart from i7 being 6C/12T vs the i5 6C/6T isn't there any other difference between them?

Regarding the Tesla GPU you mean in addition to the Nvidia Quadro card also putting in a Tesla GPU would improve performance?
You can find a list of supported GPUs here:

http://www.ansys.com/-/media/ansys/...rator-co-processor-capabilities-181.pdf?la=en

I didn't see a Quadro P4000 on this list, but I could be mistaken. Is the Quadro P4000 a Kepler? ANSYS supports the Kepler GPUs.

I have an old NVidia Quadro 5000 video card in my computer at work, and ANSYS didn't support it as a GPU. I have a NVidia Tesla K20c GPU in the computer in addition to the video card that ANSYS uses for calculations. I think we paid about $1500 for it back in the day.
 

Kurt Clark

n00b
Joined
Mar 14, 2017
Messages
11
I'd look around and see if anyone has done tests on scaling/chipset. Solidworks will like the fastest per-core in general (the 8700K), but Ansys seems to scale on cores.

https://www.pugetsystems.com/labs/articles/ANSYS-Mechanical-Fluent-Benchmark-Analysis-883/

Actually, this suggests a used dual-cpu Xeon with gobs of memory is the way to go for Ansys. (Even if it's ddr3)
I'll mostly be playing around with Ansys AIM, and yes I was aware of Intel Xeon workstations and that Ansys recommends them for Fluent, CFX and Mechanical with huge amount of memory and I would have loved to go for it but my budget was around USD2500 and I don't think those systems would have been possible in that budget.

Also I had come across this link
http://www.boxx.com/solutions/manufacturing-product-design/ansys
which shows some Ansys workstations with core i7 processors so that further gave me some confidence.
 

Derfnofred

Gawd
Joined
Dec 11, 2009
Messages
599
I think my concerns are out there (memory speed), and glad Engr62 added the bit about hyperthreading. Didn't know about that and leaves me curious about COMSOL's results there. The rest of the build looks solid. Not sure if EVGA is still doing their PSU sales, but I did just pick up a 650 GQ from them for $50, which saves you a penny. Edit--there's also that 960 GB SSD deal in the hotdeals that'd be worth looking at.

Also make sure you've got lots of data redundancy built in (offsite, ideally).

Just to add (I got curious, so I looked up in COMSOL's knowledge base) -- since most FEA/FEM programs have similar computational limitation (mesh solvers at the end of the day), it backs up what Engr62 wrote that the i5 will probably work just as well as an i7. And Solidworks benefits most from fast cores, so you're not likely to see a speedup from going i7 vs i5 here either. The 8600k and some judicious OC'ing does look your sweet spot in that architecture.

Hyperthreading
COMSOL does currently not benefit from hyperthreading. This means that by default, COMSOL will use as many threads as there are physical CPU cores on the system. The result is that if hyperthreading is active, the Windows Task Manager will show at most 50% CPU utilization (for the COMSOL process) when COMSOL is running. This is expected when hyperthreading is activated. Turning hyperthreading off will not increase COMSOL performance; we recommend that hyperthreading is used if available in order to get reasonable performance for other applications when COMSOL is running.
 
Last edited:

Kurt Clark

n00b
Joined
Mar 14, 2017
Messages
11
Just to add (I got curious, so I looked up in COMSOL's knowledge base) -- since most FEA/FEM programs have similar computational limitation (mesh solvers at the end of the day), it backs up what Engr62 wrote that the i5 will probably work just as well as an i7. And Solidworks benefits most from fast cores, so you're not likely to see a speedup from going i7 vs i5 here either. The 8600k and some judicious OC'ing does look your sweet spot in that architecture.
Thanks for this amazing information Derfnofred.
So now I have 4 choices
i5 8600k - this seems to be the sweet spot
i7 8700k - more threads but doesn't seem to matter in this scenario (would the more threads come into use in certain tasks??)
i7 7800x - more memory channels and support for upto 128GB (possibly more suited to bandwidth hungry simulation tasks)
ryzen7 1800x - this would probably have been best suited because of the 8cores. I'm not really sure how the simulation / CAD softwares are tuned to ryzen processors.

What do you think of the Ryzen 7 in this scenario?

PS. it also seems that the Ryzen 7 1700 is a better buy than the 1700x and the 1800x all seem to be the same silicon. So they say its best to overclock the 1700 to 1800x speeds.
 
Last edited:

Derfnofred

Gawd
Joined
Dec 11, 2009
Messages
599
Hi, Kurt,

In all the benchmarking I've seen, Ryzen in any form hasn't been the creme de la creme in either SW nor Ansys, so even in the budget role, I'd be buying the 8600K. And these types of packages are less likely to be finding some kind of multithreaded speedup in the near future (especially not Ansys, which is already supercomputer-scalable), which makes the extra physical cores of the Ryzen 7 less beneficial than the extra speed of the Intel cores.

So for me, it's either the 8600K as spec'd above or the 7800x with a 4x8 gb memory arrangement (make sure the motherboard supports it too!). A case for the 8700K could be made on non-SW and non-Ansys applications (and disabled at least for the latter), but the more I look the more I see people saying HT is neutral-to-detrimental in scientific software.
 
Top