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

CPU& ram config to achieve acceptable simulation time

Register Blogs Members List Search Today's Posts Mark Forums Read

Reply
 
LinkBack Thread Tools Display Modes
Old   April 25, 2013, 04:09
Default CPU& ram config to achieve acceptable simulation time
  #1
New Member
 
Ariel
Join Date: Apr 2013
Posts: 5
Rep Power: 4
Blurp2x1 is on a distinguished road
Dear All,
thank you for reading this.

introduction:
I am have an engineering background (themodynamic) and do quite some natural ventilation study's. Recently we get larger and larger demands in the usage of cfd-scale. Now we have master planning projects that require airflow simulations (isothermal) on small neighborhoods to city scale. The small neighborhoods require up to 10 to 20 Million cells. the city analysis go up to 40 to 50 million cells.

Question:
Due to these two work scopes new investments are needed in our company.
Basically we cant afford to wait 4 to 5 days till a simulation is finished. Is there a good relationship in terms of nmbr of meshing cells , cpu+ram and computation time ( we use k-epsilon , Reynolds and Navier stokes equations mostly)


Advice on this field would be real valuable and insightful. If more info is needed for a good advice i will give so. Thank you for thinking with me on this.
Blurp2x1 is offline   Reply With Quote

Old   April 25, 2013, 09:02
Default
  #2
Senior Member
 
Erik
Join Date: Feb 2011
Location: Earth (Land portion)
Posts: 486
Rep Power: 9
evcelica is on a distinguished road
The software you plan to use, and your licencing for how many cores you will be able to run will help us give you recommendations.
evcelica is offline   Reply With Quote

Old   April 25, 2013, 09:18
Default
  #3
New Member
 
Ariel
Join Date: Apr 2013
Posts: 5
Rep Power: 4
Blurp2x1 is on a distinguished road
The software will be star ccm or fluent from ansys
We will license for a powerlicense probably which gives no restrictions on the amount of cores ( in the case of star ccm.) but only if we need it.

The goal is to get the simulation time done within 2 days or three days. Today's discussion we made clear we will focus on the 10 to 20 M cells max.

Again, some guidance on the specs needed to establish this would be a real help.

KR B

Last edited by Blurp2x1; April 26, 2013 at 02:00.
Blurp2x1 is offline   Reply With Quote

Old   April 26, 2013, 02:50
Default
  #4
New Member
 
Ariel
Join Date: Apr 2013
Posts: 5
Rep Power: 4
Blurp2x1 is on a distinguished road
Would [ Nmbr of cells / cores] *42= hrs of simulation time be a good one?

assuming the we dont have a difficult solver and enough Ram on board?

example : (15 million cells / 16 cores) *42 = 39 hrs
Blurp2x1 is offline   Reply With Quote

Old   April 27, 2013, 12:38
Default
  #5
Senior Member
 
Charles
Join Date: Apr 2009
Posts: 179
Rep Power: 9
CapSizer is on a distinguished road
Quote:
Originally Posted by Blurp2x1 View Post
Would [ Nmbr of cells / cores] *42= hrs of simulation time be a good one?
assuming the we dont have a difficult solver and enough Ram on board?
example : (15 million cells / 16 cores) *42 = 39 hrs
Well, it doesn't sound unreasonable, but we are talking very approximate numbers here. What you haven't stated is if you are using a transient solver with heat transfer switched on or not. A simple, steady state, no energy equation solution would be much faster than your number above. You've mentioned isothermal? But transient or steady makes more than an order of magnitude difference, and if you are doing transient, your total time that you are simulating is the determining parameter.

16 cores is not a very good number, from a software license point of view. It is too small to make good use of a power session license, and Ansys HPC goes (I think) 8-32-128-512. From that point of view, you might as well have 32 cores, because software costs totally swamps the hardware expenses.
CapSizer is offline   Reply With Quote

Old   April 28, 2013, 21:54
Default clearity
  #6
New Member
 
Ariel
Join Date: Apr 2013
Posts: 5
Rep Power: 4
Blurp2x1 is on a distinguished road
Thank you for the clarity.

Yes our work will be 90% steady state + the calcs will be wind driven without any heat solving in the RANS calcs. The goal is to get an understanding which type of pc is needed for our work. We want to invest in computing hardware that will do the projects that has 20M cells within 2 or three days.

Budget wise we can invest for computer + licensing till +/- 60,000 US Dollar.
But if really needed it is stretchable.

The next step will be (this week) to get familiar if we need to get a cluster (wich sounds a bit alien to me). It would be real helpful if people can pinpoint if there are plug/n/play pc's on the market, or if it's a real process to aquire such.
We have a contract with Dell, but if they don't have pc's in this range other producers have to be approached.

It would be great if a few words could be shared with me. This guides me in the correct direction. But allready thanks for the advice that is given.

warmest regards ,

ANgelo
Blurp2x1 is offline   Reply With Quote

Old   April 29, 2013, 01:47
Default
  #7
Senior Member
 
Charles
Join Date: Apr 2009
Posts: 179
Rep Power: 9
CapSizer is on a distinguished road
In your position I would look at a Dell workstation with dual Xeon E5-2650 CPU's and 16 X 4 GB 1600MHz memory modules. That will give you 16 cores, and will easily meet your time targets. If you buy Ansys HPC up to 32 cores, it will mean that you cannot fully utilise the license, but at least you can fully utilise the hardware. If you require more performance, there is the option of adding another similar workstation and networking them together. You could consider hooking them up point to point with Infiniband network cards, which will save the considerable expense of an Infiniband switch. Get a network professional to sort that connection out for you.

Last edited by CapSizer; April 30, 2013 at 04:02.
CapSizer is offline   Reply With Quote

Old   April 30, 2013, 03:32
Thumbs up
  #8
New Member
 
Ariel
Join Date: Apr 2013
Posts: 5
Rep Power: 4
Blurp2x1 is on a distinguished road
Capsizer,
Thank you for the insightful comments and help you shared with me.
THx
Blurp2x1 is offline   Reply With Quote

Reply

Thread Tools
Display Modes

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 On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
pimpleFoam: turbulence->correct(); is not executed when using residualControl hfs OpenFOAM Running, Solving & CFD 3 October 29, 2013 09:35
same geometry,structured and unstructured mesh,different behaviour. sharonyue OpenFOAM Running, Solving & CFD 13 January 2, 2013 23:40
How to find a good time scale strategy? StefanG CFX 19 June 8, 2012 08:41
Time step in transient simulation shib FLUENT 0 June 17, 2010 13:07
Long time CHT transient simulation time step.... JP CFX 0 May 9, 2008 03:36


All times are GMT -4. The time now is 04:53.