6 Replies Latest reply on Apr 26, 2009 3:39 AM by Prashanth Srinivasan

Simulation benchmark testing and performance

Rick Nagel Apr 23, 2009 11:13 AM

We have recently upgraded our computers. One of the expectations was that analysis computation time would decrease. To date I have not seen a decrease.

Is there some sort of "switch" like the "3gig" thing I need to employee to make this happen? Any insight would be appreciated. The change in computers was substantial and is outlined below but again no performance gain.

OLD PC:
Dell Laptop M6300
XP 32 bit OS
a single dual core 2.6 GHz processor
3 GB of addressable memory using the "3 gig switch"
Video card NVIDA Quadro FX 3600M

NEW WORKSTATION:
Dell T7400
XP 64 bit OS
dual quad core Xeon processors at 3.0 GHz
8 GB of RAM
Video card NVIDA Quadro FX 4800

thanks in advance.

P.S. When reviewing the task manger during an analysis most of the cores never seem to max out and typically run at 25% capacity or less.

Simulation benchmark testing and performance

Prashanth Srinivasan Apr 23, 2009 4:29 PM (in response to Rick Nagel)

Hi joe, I've faced similar problems at my company too & I may be able to help you out a bit here.

Can you tell me typically how many elements and degrees of freedom are involved in your analysis? Also which solver are you using - Direct Sparse or FFEPlus (Iterative)?
Like Show 0 Likes(0)

Actions
- Simulation benchmark testing and performance
  
  Rick Nagel Apr 23, 2009 6:33 PM (in response to Prashanth Srinivasan)
  
  A "typical" analysis may be 106797 nodes, 65849 elements, and 319536 DOF. The solver type is usually FFEPlus.
  
  thank you
  
  Like Show 0 Likes(0)
  
  Actions
  - Simulation benchmark testing and performance
    
    Prashanth Srinivasan Apr 24, 2009 11:54 PM (in response to Rick Nagel)
    
    joe g wrote:
    
    A "typical" analysis may be 106797 nodes, 65849 elements, and 319536 DOF. The solver type is usually FFEPlus.
    
    thank you
    
    joe, from my experience the FFEPlus does not benifit significantly from upgrading to quad core from dual core. This is because of two reasons -
    1. the speed of each processor stays the same
    2. the FFEPlus solver itself is not designed to make maximum usage of available processor speed
    
    That probably explains why you haven't seen an improvement in performance by going to quad-core - I assume that on your old dual-core system, the solver peaked at 50% for each processor.
    
    The best solution in this case is not to go for more number of processors, but to go for a FASTER processor (meaning more speed per processor). The best option in this case would be the Dual Core EXTREME processor, by which you will get significant speed improvement for the same processor usage.
    
    That's my 2 cents
    
    Like Show 0 Likes(0)
    
    Actions
    - Simulation benchmark testing and performance
      
      Prashanth Srinivasan Apr 26, 2009 3:39 AM (in response to Prashanth Srinivasan)
      
      Alternatively, you might also try overclocking your CPU speed from 3.0 GHz to higher value and check if there is significant improvement
      
      Like Show 0 Likes(0)
      
      Actions
Simulation benchmark testing and performance

1-YJ1IM Apr 23, 2009 8:01 PM (in response to Rick Nagel)

I think the direct sparse solver is the only one that can really take advantage of multi processors.
Like Show 0 Likes(0)

Actions
Simulation benchmark testing and performance

Anna Wood Apr 24, 2009 3:22 AM (in response to Rick Nagel)

You do not need the 3Gb switch for a x64 bit OS. The 3Gb switch is for those with x32 bit OS's to be able to have a bit more memory to work with in your programs as opposed to the OS.

Cheers,
Like Show 0 Likes(0)

Actions

Go to original post