WorkStation for Hobbyist-Researcher

ansestor · May 16, 2018, 04:29

Dear all,
I am looking for configuration to work mostly with ANSYS Mechanical APDL and maybe in the future with LAMMPS software. Currently I am about to finish my PhD. At university my PC is an old i5-3470 with only 8GB. If needed I have option to solve my numerical tasks in a cluster but most of my models do not exceed 1mln elements. I rather work with smaller models 100k-500k elements. Usually I write APDL code for model and run calculations with different material properties or geometry dimensions. That's why I would like to have sth fast enough to perform this tasks and not to wait few days to finish solving all cases.
My budget isn't any big, though. Around 1000E. So, I am thinking about Ryzen 7 or i7-8700. After doing some reading about pros and cons I came up with those two configurations:
AMD
CPU R7 2700
MOBO MSI PC MATE
RAM 32GB 3000 Corsair
SSD Plextor 256GB Heat Sink
HDD Toshiba P300 1TB
GPU 1050Ti 4GB
PSU Seasonic Focus 550 Gold

INTEL
CPU 8700
MOBO AsRock B360 PRO4
rest same as above

Any input will be appreciated.

flotus1 · May 16, 2018, 05:07

While the new generation of Ryzen Processors is a small step in the right direction, it still can not quite compete with Intels mainstream CPUs in terms of performance per core.

So what I would recommend instead:
Intel I5-8600k
Some Z370 motherboard
2x16GB DDR4-3200 or faster

You can also get an I7 instead, but SMT is mostly useless for this kind of application.
I highly recommend a Z370 motherboard. There are some decent ones at around 100$ (ASRock for example) and you won't be able to use faster RAM with any of the lower tier chipsets.

ansestor · May 16, 2018, 05:15

Thank you for fast response.
Well you have good point with this Z370 chipset. But instead of i5-8600K i would rather go with i7-8700 and AsRock Z370 Pro4 (first I found and rather cheap option for MOBO). Intel has also one advantage for me. It not requires dedicated GPU. So i would use Intel integrated graphic and spent money for better RAM.

Let's say my setup now looks as follows:
CPU 8700 (6core 12 thread)
MOBO AsRock Z370 Pro4
RAM 32GB 3000 Corsair
SSD Plextor 256GB Heat Sink
HDD Toshiba P300 1TB
PSU Seasonic Focus 550 Gold (already bought week ago).

What do you think? Leave like this or try to strech my budget a bit and take 8700K?

flotus1 · May 16, 2018, 06:19

I see no compelling reason to spend more money on an I7, especially one with a locked multiplier like the I7-8700. What do you expect from it?

ansestor · May 16, 2018, 06:31

Wouldn't be helpful to have 6core - 12threads in comparison to 6C/6T.
Aren't this extra 6T allow to make more iterations during calculations?

flotus1 · May 16, 2018, 08:22

Like I said, SMT is mostly useless for the tasks involved here. I still recommend switching it off for CFD/FEA workstations.
The extra money will be better spent on a decent CPU cooler and faster RAM.

ansestor · May 16, 2018, 08:35

So basically in ANSYS calculations is the same situation as in CAD software. It's better to have strong care instead of bunch of weak ones. Am I understand this correctly?
Unfortunately I am not a specialist in terms of hardware. Until now I had only my mentioned above i5 workstation and I knew that it's my only option.In order to use cluster at my University I prepared models (mesh, boundary condition s etc) and friend of mine did the magic with sending files with correct settings.
That's why I may sound like an ignorant

flotus1 · May 17, 2018, 05:05

The requirements for CFD/FEA solvers and CAD may seem similar at first, but they are different.
Many workloads in CAD are not inherently parallel, so they can not use many cores. Hence a CPU with a small amount of fast cores is preferred.
CFD/FEA solvers on the other hand are inherently parallel so they could benefit from a larger amount of cores. The issue here is memory bandwidth. 6 threads as in an I5-8600k are usually enough to saturate the memory bandwidth its two memory channels can provide. So more cores would not help. On top of that there is parallelization overhead. it increases with higher thread count and smaller model size.
SMT or Hyperthreading as Intel calls its implementation is a different story. It basically tries to offload some computations to unused parts of a core while other parts are busy. This would require additional data from memory and is thus ineffective when the memory bandwidth is already saturated. In addition to that, It further increases parallelization overhead (twice the amount of threads) and can even increase solution times.
With some very fast RAM you can sometimes get a slight performance increase from SMT, but in my opinion it is not worth it considering the downsides - higher hardware cost, more licenses used depending on the licensing policy of your software and lower performance in some cases.
To sum things up: if I had to buy one of Intels mainstream CPUs now, it would definitely be the I5-8600k.

ansestor · May 17, 2018, 08:41

@flotus1. Thank you for your complex explanation. I won't say that I understand every technical detail but I get your point.
I'll stick to the Intel platform with i5-8600k.
Other parts are:
MOBO ASRock Extreme 4
RAM 32GB 3200MHz - whatever manufacturer will be cheaper
SSD Plextor M8Pe 256GB
no GPU or in should say Intel 630HD.

Hopefully I will be able to add another 32GB of RAM in e near future.
Again, thank you Alex for your help and useful information about FEA solver technology.

pibil1 · May 19, 2018, 20:23

Why not i7 5820k or 7800x recommendations anymore Alex?

Quote:

Originally Posted by flotus1

While the new generation of Ryzen Processors is a small step in the right direction, it still can not quite compete with Intels mainstream CPUs in terms of performance per core.

So what I would recommend instead:
Intel I5-8600k
Some Z370 motherboard
2x16GB DDR4-3200 or faster

You can also get an I7 instead, but SMT is mostly useless for this kind of application.
I highly recommend a Z370 motherboard. There are some decent ones at around 100$ (ASRock for example) and you won't be able to use faster RAM with any of the lower tier chipsets.

flotus1 · May 20, 2018, 06:54

5820k is used and needs to be overclocked to beat a 8600k. Not everybody wants that.
7800x is significantly more expensive considering the platform costs and should also be tweaked a little to use its full potential.

May 16, 2018, 04:29	WorkStation for Hobbyist-Researcher	#1
ansestor New Member Tomasz Join Date: May 2018 Posts: 5 Rep Power: 8	Dear all, I am looking for configuration to work mostly with ANSYS Mechanical APDL and maybe in the future with LAMMPS software. Currently I am about to finish my PhD. At university my PC is an old i5-3470 with only 8GB. If needed I have option to solve my numerical tasks in a cluster but most of my models do not exceed 1mln elements. I rather work with smaller models 100k-500k elements. Usually I write APDL code for model and run calculations with different material properties or geometry dimensions. That's why I would like to have sth fast enough to perform this tasks and not to wait few days to finish solving all cases. My budget isn't any big, though. Around 1000E. So, I am thinking about Ryzen 7 or i7-8700. After doing some reading about pros and cons I came up with those two configurations: AMD CPU R7 2700 MOBO MSI PC MATE RAM 32GB 3000 Corsair SSD Plextor 256GB Heat Sink HDD Toshiba P300 1TB GPU 1050Ti 4GB PSU Seasonic Focus 550 Gold INTEL CPU 8700 MOBO AsRock B360 PRO4 rest same as above Any input will be appreciated.

May 17, 2018, 05:05		#8
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,427 Rep Power: 49	The requirements for CFD/FEA solvers and CAD may seem similar at first, but they are different. Many workloads in CAD are not inherently parallel, so they can not use many cores. Hence a CPU with a small amount of fast cores is preferred. CFD/FEA solvers on the other hand are inherently parallel so they could benefit from a larger amount of cores. The issue here is memory bandwidth. 6 threads as in an I5-8600k are usually enough to saturate the memory bandwidth its two memory channels can provide. So more cores would not help. On top of that there is parallelization overhead. it increases with higher thread count and smaller model size. SMT or Hyperthreading as Intel calls its implementation is a different story. It basically tries to offload some computations to unused parts of a core while other parts are busy. This would require additional data from memory and is thus ineffective when the memory bandwidth is already saturated. In addition to that, It further increases parallelization overhead (twice the amount of threads) and can even increase solution times. With some very fast RAM you can sometimes get a slight performance increase from SMT, but in my opinion it is not worth it considering the downsides - higher hardware cost, more licenses used depending on the licensing policy of your software and lower performance in some cases. To sum things up: if I had to buy one of Intels mainstream CPUs now, it would definitely be the I5-8600k. Last edited by flotus1; May 17, 2018 at 10:34. Reason: forgot words and stuff

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Suggestions for StarCCM+ Workstation configuration	ifil	Hardware	15	October 30, 2018 06:09
Abysmal performance of 64 cores opteron based workstation for CFD	Fauster	Hardware	8	June 4, 2018 11:51
Recommendation to buy a workstation 2018	sherlock	Hardware	3	March 6, 2018 09:41
Dual cpu workstation VS 2 node cluster single cpu workstation	Verdi	Hardware	18	September 2, 2013 04:09
Pre- and Post workstation - need inputs	larsenmm	Hardware	17	February 13, 2013 05:16

May 16, 2018, 05:07		#2
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,427 Rep Power: 49	While the new generation of Ryzen Processors is a small step in the right direction, it still can not quite compete with Intels mainstream CPUs in terms of performance per core. So what I would recommend instead: Intel I5-8600k Some Z370 motherboard 2x16GB DDR4-3200 or faster You can also get an I7 instead, but SMT is mostly useless for this kind of application. I highly recommend a Z370 motherboard. There are some decent ones at around 100$ (ASRock for example) and you won't be able to use faster RAM with any of the lower tier chipsets.

May 16, 2018, 05:15		#3
ansestor New Member Tomasz Join Date: May 2018 Posts: 5 Rep Power: 8	Thank you for fast response. Well you have good point with this Z370 chipset. But instead of i5-8600K i would rather go with i7-8700 and AsRock Z370 Pro4 (first I found and rather cheap option for MOBO). Intel has also one advantage for me. It not requires dedicated GPU. So i would use Intel integrated graphic and spent money for better RAM. Let's say my setup now looks as follows: CPU 8700 (6core 12 thread) MOBO AsRock Z370 Pro4 RAM 32GB 3000 Corsair SSD Plextor 256GB Heat Sink HDD Toshiba P300 1TB PSU Seasonic Focus 550 Gold (already bought week ago). What do you think? Leave like this or try to strech my budget a bit and take 8700K?

May 16, 2018, 06:19		#4
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,427 Rep Power: 49	I see no compelling reason to spend more money on an I7, especially one with a locked multiplier like the I7-8700. What do you expect from it?

May 16, 2018, 06:31		#5
ansestor New Member Tomasz Join Date: May 2018 Posts: 5 Rep Power: 8	Wouldn't be helpful to have 6core - 12threads in comparison to 6C/6T. Aren't this extra 6T allow to make more iterations during calculations?

May 16, 2018, 08:22		#6
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,427 Rep Power: 49	Like I said, SMT is mostly useless for the tasks involved here. I still recommend switching it off for CFD/FEA workstations. The extra money will be better spent on a decent CPU cooler and faster RAM.

May 16, 2018, 08:35		#7
ansestor New Member Tomasz Join Date: May 2018 Posts: 5 Rep Power: 8	So basically in ANSYS calculations is the same situation as in CAD software. It's better to have strong care instead of bunch of weak ones. Am I understand this correctly? Unfortunately I am not a specialist in terms of hardware. Until now I had only my mentioned above i5 workstation and I knew that it's my only option.In order to use cluster at my University I prepared models (mesh, boundary condition s etc) and friend of mine did the magic with sending files with correct settings. That's why I may sound like an ignorant

May 17, 2018, 08:41		#9
ansestor New Member Tomasz Join Date: May 2018 Posts: 5 Rep Power: 8	@flotus1. Thank you for your complex explanation. I won't say that I understand every technical detail but I get your point. I'll stick to the Intel platform with i5-8600k. Other parts are: MOBO ASRock Extreme 4 RAM 32GB 3200MHz - whatever manufacturer will be cheaper SSD Plextor M8Pe 256GB no GPU or in should say Intel 630HD. Hopefully I will be able to add another 32GB of RAM in e near future. Again, thank you Alex for your help and useful information about FEA solver technology.

May 20, 2018, 06:54		#11
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,427 Rep Power: 49	5820k is used and needs to be overclocked to beat a 8600k. Not everybody wants that. 7800x is significantly more expensive considering the platform costs and should also be tweaked a little to use its full potential.