workstation performance / analysis time

[rmckinley]

Does anyone know what I might stand to gain from a quad-core or dual quad-core system setup? Some of my analyses take an excessive length of time to complete and I am looking at upgrading. I have been told that the biggest improvement comes from increased RAM, however, I suspect that along with this, I should consider multiple processors or cores and a fast hard drive configuration (for example, SAS 15,000RPM perhaps in a RAID0). Or does anyone know where I can get benchmark info on new workstations as they pertain to Pro/MECHANICA? I'm really after how much time I could actually save by upgrading.

 

[Mikhail33]

Try http://www.proesite.com/

 

[mediumsliced]

AFAIK, Pro/E & Mechanica will not utilize multiple cores. Just try this, run an analysis, launch your task manager and look at the CPU utilization. For a dually setup, you should see your CPU being loaded up to 50% max.

Anyway, there's an option of the BIOS of your box to disable multi-cores or HT (that doesn't work too).

The only performance improvement I can foresee with multi-core boxes is with running other apps simultaneously.

HTH.

 

[Kaz Z06]

Pro/Mechanica WILL take adavantage of multiple processors. You need to set an environment variable:


MECH_NUM_THREADS = number of CPUs


I have been using this setting for years, and most recently for the newer "multi-core" processors,and it will make a differnce in total run time for many problems. Performance will vary based on the type of analysis you are solving and how much RAM/Disk you have. I have included portions of a ".rpt" filethat document the number of processors set and recognized for use. At the very top you'll notice the following:


------------------------------------------------------------


Mechanica Structure Version K-03-42:spg


Summary for Design Study "QJB2500_24mm_30mm_2300kg"


Mon Nov 20, 2006 08:29:20


------------------------------------------------------------


Run Settings


Memory allocation for block solver: 250.0


Parallel Processing Status


Parallel task limit for current run: 2


Parallel task limit for current platform: 64


Number of processors detected automatically: 2





A the end of the report you see something like this:


------------------------------------------------------------


Memory and Disk Usage:


Machine Type: Windows NT/x86


RAM Allocation for Solver (megabytes): 250.0


Total Elapsed Time (seconds): 458.01


Total CPU Time (seconds): 626.41


When the "Total Elapsed Time" is less than "Total CPU Time", you know your computer took advantage of multiple processors. I hope this helps out. Remember, you may not see much of a differnece and you may need to run the same analysis with/without multiple CPUS enabled to understand when it will work best. Simple, short runs will make no difference at all, as will problems that may be "I/O bound" due to insufficient RAM.


Regards
Edited by: Kaz Z06

 

[mediumsliced]

Thanks for the tip. Now if only Pro/E has something like that...

 

[rmckinley]

Thanks everyone. I know that proe won't take advantage of multiple processorsor cores but I also know that Mechanica definitely does (and, afaik, by default it uses all available).


I would like to know if anyone has any data or even anecdotal evidencethat illustratestime savedby using any of the powerful multi-coreand/or RAIDhardware configurations thatare available today.


I currently run Mechanica on a 3.0GHz Xeon with 2GB RAM and an ultraATA drive. I set my solram to 1000. Sometimes it's justtoo slow. Enabling hyperthreading causes Mechanica to detect two processors but there is no differencein performance (as we are all aware - HT doesn't reallyprovide much improvementand sometimes slows things down). Is it likely that this configuration would still have significant enough I/O to benefit from a SAS drive or RAID configuration? And how much do I gain from 2 or 4 or 8 cores? Would it be worth spending the $4-8k that these machines cost? It would also be nice if I could keep working on other thingswhile the solver is running. Currently, it pegs my processor and hard disk making it frustrating to attempt to use the computer while the solver is running (and, of course, using the machine while the solver is running makes the solver take even longer to finish). I would even consider running multiple simulations simultaneously ifnew hardware can make itfeasible.

 

[magneplanar]

Mediumsliced,

I think this config option allows for multiple threads like the mechanica one:

cpus_to_use *1

cheers...

M

 

[mediumsliced]

Thanks, tried it out. Does anybody else use this option? I tried turning this on, created a pattern feature with loads of members, edited and then did a regen, didn't take up 50% like before setting the option but actually went up to ~70%.