CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > General Forums > Hardware

CFD Hardware setup by small FEA benchmark. Would it make sense?

Register Blogs Community New Posts Updated Threads Search

Reply
 
LinkBack Thread Tools Search this Thread Display Modes
Old   May 8, 2021, 11:18
Default CFD Hardware setup by small FEA benchmark. Would it make sense?
  #1
New Member
 
Join Date: May 2015
Posts: 2
Rep Power: 0
Twenty is on a distinguished road
Dear all, good afternoon.

It's quite long time I read trough this website; this is my first post and before to start with my (boring) stuff I'd say thanks to everybody for the contributions, experiences and information shared.

After reading many posts about hardware configuration for CFD I think I partially cleared some confusion I had in my mind, although not being an expert of this field, still leaves me several thoughts.
I want to build a personal machine for running CFD and my idea would be to start learning OpenFoam, so, for the moment I'd ask some suggestion on hardware stuff and I'd like to share some examples.
I am a Structural Engineer; I did perform some basic CFD jobs in my company (with Altair AcuSolve) but just on specific and limited subjects.
CFD Is officially a new field for me; with lots to learn and limited time, I just need of a "decent" machine, not huge, not too expensive, but powerful enough to satisfy the needs of a beginner but also, to be reliable for other computational (non fluid) work. Let say my budget is max 1500 Pounds.

Thinking about the memory bandwidth as a key parameter and staying within the mainstream cpus sector, (many thanks gents, for all the post I read here) my brain got stucked to the Intel i7 9800X.
It looks fast in terms of clock speed, overclockable, 4 mem. channels, good RAM frequency and 8 cores seem a good balance... Wow, let's go !... but then, I said.. wait a moment, let me perform some basic benchmark using our machines in the office, to get some number.. ok ok... not by CFD unfortunately, sorry, but if I compare hardware performances of different machines by the same FEA problem, by the rule of thumb.. more or less I'll learn something useful for my future adventure in CFD.

So, the attached picture (I hope it's exhaustive in terms on data and specs) shows how 3 machines perform the same task; more in detail, the job was a static nonlinear analysis ran in Abaqus Standard (Implicit) with 1,2,4,8,16,24 cores (no gpu).
The model was a mechanical assembly made up of about 350000 elements (mainly tetra 2nd order C3D10 + some 1st order C3D8I) divided in 5 components, all in contact together under a couple of loading conditions, with different contact settings (soft and hard), running in two steps. (sorry for these details, just to describe a bit more the case analysed).
It is a medium/small usual problem we have to face every day; it does not kill any machine, it runs "quickly" and I believed it was a good example to extract some thoughts.

All the machines have always been used as "out of the box".. from DELL, straight to the office, no tests, no tunings, only with Windows OS .. (so, the judgement is based on these assumptions...)
In my opinion the Workstation laptop does pretty well even at 8 cores; the two old "A" and "B" machines scale slightly better in percentage if you look how steep are the line up to 8 cores. Now, both "A" and "B" are at the end of their life, gpus both KO, and "B" has "lost" 8 cores apparently ...

So, my point, to be transferred to my personal "desired" CFD machine was... mmm... If the i7 did this job in dual channel with that (strange) memory config... (it has a NVME ssd as disk...)
How could the i7 9800x be? quad channel.. better memory configured... would it be... a bit faster? faster? much faster? I would hope definitely better, but how to estimate it?
FEA uses probably the computational resources in a slightly different way than CFD but this is what I have to get some comparison and I guess it is anyway helpful about hardware behaviour.
Another point I'd like to clear is how to compare cpus just by their specs? For instance... A ryzen 7 3800X or 5800X have the same L1 cache of the i7 9800X, but the L2 cache of the i7 is twice the L2 cache of the Ryzen.. Again, the L3 cache of the Ryzen is now twice the amount of the L3 cache of the i7. .. Is it related to the different memory layout ? (dual vs quad channel) And, would this afflict/improve performances?
Other aspect.. a Ryzen 7 3800X (for example) runs 3200 MHz RAM, while the i7 runs 2666 MHz RAM. The total bandwidth is in favour of i7 but different caches and memory speeds complicate my personal understanding on "what is the best"...

I ended up the patience of several friends asking about these (maybe useless?) detailed arguments but we (me, for sure) want the best, especially when investing money in hardware that loses value pretty quickly. For my personal research activity/fun I believe any i7, Ryzen7 or whatever cpu with 8 cores max. could be enough to get nice job initially, but we know, learning day by day and making more complex models reaches the point where it is easy to say.. "if I'd have bought that cpu instead of this one... " ... and within 1500 pounds I think the choice of the right CPU is quite important.
By the way.. with a poor knowledge of hardware details and high curiosity, for me, it is also easy to focus the attention on cache, mem channels and all sort of details that give a touch of "mystery" to the built..

Any thoughts or suggestion would be really appreciated. Thanks for reading,
20
Attached Images
File Type: png cpus_comparison.PNG (84.4 KB, 26 views)
Twenty is offline   Reply With Quote

Old   May 10, 2021, 07:24
Default
  #2
Member
 
dab bence
Join Date: Mar 2013
Posts: 48
Rep Power: 13
danbence is on a distinguished road
The best advice I can give to a beginner at CFD is not spent too long on the finer details of hardware selection. If you follow the general principles that flotus1 set out here ( General recommendations for CFD hardware [WIP] ), you can achieve a decent machine for CFD.

When you are starting out in CFD, there is no question that your time is best spent learning how to optimize the cell count, timestep, simulated time duration and convergence criteria. I have managed to optimize models to give 100x or 1000x speedup. Selecting between two hardware configurations of similar price might change the runtime by 20% (say), which would be swamped by a tiny change in cell size.
danbence is offline   Reply With Quote

Old   May 12, 2021, 11:03
Default
  #3
New Member
 
Join Date: May 2015
Posts: 2
Rep Power: 0
Twenty is on a distinguished road
Thanks for your suggestions/point of view danbence.
20
Twenty is offline   Reply With Quote

Reply


Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
CFD Salary CFD Main CFD Forum 17 January 3, 2017 18:09
CFD Design...The CFD Future John C. Chien Main CFD Forum 20 November 20, 2015 00:40
PC vs. Workstation Tim Franke Main CFD Forum 5 September 29, 1999 16:01
goal of CFD MYM,Rameez Main CFD Forum 7 August 16, 1999 01:12
Which is better to develop in-house CFD code or to buy a available CFD package. Tareq Al-shaalan Main CFD Forum 10 June 13, 1999 00:27


All times are GMT -4. The time now is 07:20.