Increase Processing Speed in FEA MES

[LongLe]

Hey guys..me agian

So I tried to compare my new machine to my old one....
I was able to run the Piston MES

Old Machine completed with a time of 49 min 10 seconds. Constant CPU Usage 55-61%

New Machine completed with a time of 44 min 9 seconds.
Constant CPU Usage 15-20%

Is there something im doing wrong with my workstation configurations? The only option I see is to select the number of processors right before I begin the analysis. Is there a setting somewhere that im not seeing. I thought by buying faster hardware the system would process the MES faster...

My Specs are
CPU: Intel Core i7 920
Memory: OCZ 12GB DDR3 1600mhz
Hard-Drive: 3x Western Digital 320GB in RAID5 Configuration
Video-Card: ATI FireVL 5100
Powersupply: OCS 700watt
OS: Windows XP 64bit SP2

Old Machine:
CPU: Intel Pentium 4 3.40 GHZ
Memory: DDR2 667mhz
Hard-Drive: 1 Western Digital 320GB
Power-Supply: 450 watt
OS: Windows XP 32bit SP3

 

[GBor]

How much RAM did you have in the old machine? And, did you use the exact same model (same orientation to the global axes, same element types and numbers of nodes, etc.)

 

[johnhors]

The clock speeds of CPU cores have not increased for a few years now. Your old machine at 3.4 GHz is still about as fast as you can get. You don't quote the clock speed of your new machine, but I bet it isn't any faster. Your old machine appears to be single core (one processor) to be running at 55-61%, whilst your new machine I reckon is a quad core and hence the much lower CPU usage as it is only making use of a single core (the software not being able to take advantage of multiple cores).

Your old machine is 32bit whilst your new machine is 64bit, therefore have you installed exactly the same software on both or does the 64bit machine have a 64bit version of the software installed? If you have done the former then the 64bit machine cannot be expected to run the code much faster than the old machine!

http://www.Roshaz.com

 

[LongLe]

My new machine is Core i7 2.67 over clock 3.4ghz. I installed the 64 Bit version of Algor on it. I am running the same model. I exported from old machine to the new machine. Johnhors I see what you are saying about the software only making use of a single core. I attached a screenshot of the CPU usage using our monitoring software but it does show usage on the other 7 cores its just not as high.

 

[LongLe]

I thought by setting "numbers of processors" to all it would use all cores. I attached a screenshot of this setting. Agian thanks for all your help. I appreciate it alot :-D .

 

[GBor]

That should do it, but check the "solution" tab under "Analysis Parameters". It should default to "all", but just in case, I'm not sure which one dictates of if they are driven from the same location.

I still say that you situation is RAM related. This model is small enough that it is likely staying "in core", which means the model is solving in the RAM. As johnhors points out, your clock speed hasn't really changed between the two computers and if both computers maintain the model "in core", then the only thing affecting solution time is the bus speed. You won't see significant improvement until one of your computers goes "out of core" while the other one stays "in core" because of the quantity of RAM that you have in each machine.

 

[LongLe]

GBor,

Im sorry I didnt really understand the in core vs out of core. I assume you mean the model is small enough to be In core \ cache on the processor or out of core \ cache on ram? I run Passmark Bench Mark and my new machine is faster in everyway. (hd i/o speed, ram i/o speed, cpu power etc...)

 

[LongLe]

I had Algor build me a model gonna give a test run. They said that the cpu power all depends on the model that you run. Algor said that this model that they made runs @ 100% CPU usage on their dual core system. I check the "solution" tab under "Analysis Parameters" as Gbor advise and Ill see of EVGA can help me bump up my speed on my board.

 

[johnhors]

I'm not convinced that it is running in any form of multi-processor mode! If you fire up task manager, go to the processes tab and select any 32bit application (those with *32) , right click and select "set affinity" you should see all processors ticked by default. This does not mean the 32bit application is written for multiple processors! Do the same for the Algor application , it should not have *32 (if it does it is a 32bit application and not 64bit), then under select affinity, uncheck all the processors bar one, does the run time change or not? If there is no significant change, then it is not multi-processing!

http://www.Roshaz.com

 

[GBor]

Algor uses multiple processors for the analysis...at least on my system. How efficiently or effectively?

Well, I do have access to more efficient software packages that better utilize multiple processors...even one that I consider much better for MES, but Algor does use multiple processors during the analysis phase.

 

[LongLe]

johnhors,

I tried what you said and only selected 1 CPU. I ran the analysis agian for 10 minutes.

Results after 10 minutes:
Old System - 18.75%
New System - 20.63%
New System with Affinity set to use 1 Core - 15%

I attached the model incase you wanted to give it a try...

 

[GBor]

This model isn't meshed. This is just the CAD geometry. Without the exact same mesh on each model, it will be improper to compare.

 

[johnhors]

My point was all applications will use all the processors irrespective of whether they are true multi-processing applications, but on a four core system a single processor app will max out at 25% and only a true multi-processor app will go beyond 25%.

http://www.Roshaz.com

 

[LongLe]

Here is a new screenshot of cpu useage running a model from Algor. So maybe it is the model that I process that determines how many core FEA uses? if you compare this screenshot to the previous one there is a huge difference

 

[repars]

Gbor-

Just curious what FEA program do you use that can better utilize multiple processors for MES type of analysis? NeiNastran, Nx Nastran, ANSYS, maybe ABAQUS??? Just curious because I'm looking into FEA software and I'd like to get the most bang for my bucks.

Also as a side note: LongLe that is a nice system you set up for running FEA. I don't know a great deal about Algor however I do know that they offer a choice of solvers, some of the older solvers may not take advantage of the multiprossor systems like the one you have set up. How many DOF was your model?

JP

 

[GBor]

http://www.ampstech.com

Algor's user interface is better and it's mesh control is a bit easier, but the engine of AMPS is nice. I use Algor for dynamic analysis (because I've been using it for over a decade and can pretty much run it in my sleep), Roshaz for certain linear static problems (CAD clean-up tools are impressive for the price), and AMPS for any impact/MES problems (algorithm is similar to LS-DYNA).