Re: Slow NX performance

Phenom
Phenom

Blas,

Did test as suggested but I think hypertreading is active. Not sure how to check but a colleague may have mentioned that in the past

time is roughly the same (3minutes) with a few seconds

  MODULE     PROCESS-1     PROCESS-2     PROCESS-3     PROCESS-4     PROCESS-5     PROCESS-6     PROCESS-7     PROCESS-8
  --------   -----------   -----------   -----------   -----------   -----------   -----------   -----------   -----------
  P-CASI          8.82          8.82          8.82          8.82          8.82          8.82          8.82          8.82
  ========   ===========   ===========   ===========   ===========   ===========   ===========   ===========   ===========
  TOTAL           8.82          8.82          8.82          8.82          8.82          8.82          8.82          8.82


 *** TOTAL MEMORY AND DISK USAGE STATISTICS ***

 +---------- SPARSE SOLUTION MODULES -----------+         +------------- MAXIMUM DISK USAGE -------------+
      HIWATER               SUB_DMAP        DMAP               HIWATER               SUB_DMAP        DMAP
      (WORDS)   DAY_TIME      NAME         MODULE                 (MB)   DAY_TIME      NAME         MODULE
  12427921067   10:18:59    SOL2      368  SOLVIT              663.562   10:20:39    SOL2      604  EXIT    


 *** DATABASE USAGE STATISTICS ***
 
 +------------------ LOGICAL DBSETS ------------------+   +------------------------- DBSET FILES -------------------------+
  DBSET      ALLOCATED   BLOCKSIZE        USED    USED     FILE       ALLOCATED     HIWATER       HIWATER  I/O TRANSFERRED
              (BLOCKS)     (WORDS)    (BLOCKS)       %                 (BLOCKS)    (BLOCKS)          (MB)             (GB)
 
  MASTER          5000       65536          46    0.92     MASTER          5000          46        23.000            0.443
  DBALL        2000000       65536        1243    0.06     DBALL        2000000        1243       621.500            0.613
  OBJSCR          5000        8192         289    5.78     OBJSCR          5000         289        18.062            0.022
  SCRATCH      4146713       65536        1250    0.03    (MEMFILE       146713        2308      1154.000            0.000)
                                                           SCRATCH      2000000           1         0.500            0.000
                                                           SCR300       2000000           1         0.500            0.000
                                                                                                            ==============
                                                                                                     TOTAL:          1.079
Production: NX.CAE 10.0.2.6
Development: VB.NET (amateur level !)

Re: Slow NX performance

Solution Partner Phenom Solution Partner Phenom
Solution Partner Phenom

Hello!,

Is simply, take note of the processor name you have installed in your machine (go to CONTROL PANEL > SYSTEM & SECURITY > SYSTEM), go to the vendor site (if intel go to https://ark.intel.com/, is the best place) and take note of the number of cores, this is the important value. Also you will see number of threads, is the double of cores, this is useless for NX NASTRAN, disable in the BIOS, OK?.


@selex_ct wrote:

Blas,

Did test as suggested but I think hypertreading is active. Not sure how to check but a colleague may have mentioned that in the past

time is roughly the same (3minutes) with a few seconds


 

Blas Molero Hidalgo, Ingeniero Industrial, Director
IBERISA • 48004 BILBAO (SPAIN)
WEB: http://www.iberisa.com
Blog Femap-NX Nastran: http://iberisa.wordpress.com/

Re: Slow NX performance

Phenom
Phenom

hypertreading is on

Production: NX.CAE 10.0.2.6
Development: VB.NET (amateur level !)

Re: Slow NX performance

Honored Contributor
Honored Contributor

Not being an NX Nastran guy, where would I find this preferences menu?  I'm trying to help our users out a bit.  Thanks.
@J0hnPatr1ck wrote:

 

 

ILP 64.PNGILP-64

 


 

-Dave
NX1867(if it had versions) | Teamcenter 11.6 | Windows 10

Re: Slow NX performance

Solution Partner Phenom Solution Partner Phenom
Solution Partner Phenom

Dear Dave,

This is a image from great FEMAP located at FILE > PREFERENCES > INTERFACES where the user can activate to use the ILP-64 bit NX NASTRAN solver instead the LP-64 solver that only address 8 GB RAM memory. The ILP-64 executable of NX NASTRAN can address millions of TERABYTES of RAM memory, and has 8 bytes per word, not 4.

Best regards,

Blas.

Blas Molero Hidalgo, Ingeniero Industrial, Director
IBERISA • 48004 BILBAO (SPAIN)
WEB: http://www.iberisa.com
Blog Femap-NX Nastran: http://iberisa.wordpress.com/

Re: Slow NX performance

Pioneer
Pioneer

Friends, 

 

Thank you for all of your help and support.

 

Blas, it's always good to see you involved in a discussion. You have helped me numerous times in the past and I really appreciate all of the efforts you put into this forum. We are lucky to have your expertise, thank you! 

 

I downloaded your input file and my computer ran the model in 1 min and 40 sec. I looked over the model an noticed that the Solid Elements were Linear elements and thus the file only had 241211 nodes. I added the mid-side nodes to match the original ANSYS model I created and the solution time is back up to 5+ minutes. I have attached the input file for my model to this discussion. 

 

I'm wondering if the reason ANSYS PCG solver can solve this problem with the solid parabolic element (948321 nodes) in under a minute is that PCG solver is working entirely on my computers 128 GB of RAM, while the Nx Elemental Iterative solver is using the scratch directory for the problem and simply using the computers 128 RAM for IO bandwidth? I really don't know and I am just speculating wildly however in my mind I feel like ANSYS should not be 600% faster than Nastran. Hase anyone ever tried using a RamDisk to create a virtual hard drive in memory to use as the scratch directory? I may try it tonight if not I'll see about this next week. 

 

I really appreciate all of the help everyone has provided to me if I'm missing something or someone has an idea please let me know.

 

Thanks

John

Re: Slow NX performance

Solution Partner Experimenter Solution Partner Experimenter
Solution Partner Experimenter

Hi John,

 

Just noticed your conversation about slow performance with Nastran.

 

Since it looks like you're working in FEMAP you do have the option of increasing the number of processors used for the analysis. In FEMAP if you go to the Analysis Set Manager, and "Edit" the analysis of interest, keep selecting "Next" until you get to the  Nastran Exectuve Solutions Options dialog as shown below. There you can set the number of processors used to solve the analysis - the default is 1.

 

FEMAP_Nastran_Processors.JPG

 

Hope this helps

 

David Christensen

EdgePLM

Re: Slow NX performance

Solution Partner Phenom Solution Partner Phenom
Solution Partner Phenom

Dear John,

If I run the high-order HEX20 model with 948321 nodes the solution time is 4 min, 30 sec in my old LapTop with 32 GB RAM and QUADCORE processor. Please note the TOTAL SMEM used (the I/O during solution) is around 7 GB.

 *** PARALLEL PROCESSES CPU INFORMATION (SECONDS) ***

  MODULE     PROCESS-1     PROCESS-2     PROCESS-3     PROCESS-4     PROCESS-5     PROCESS-6     PROCESS-7     PROCESS-8
  --------   -----------   -----------   -----------   -----------   -----------   -----------   -----------   -----------
  P-CASI         78.53         78.53         78.53         78.53
  ========   ===========   ===========   ===========   ===========   ===========   ===========   ===========   ===========
  TOTAL          78.53         78.53         78.53         78.53

 *** TOTAL MEMORY AND DISK USAGE STATISTICS ***

 +---------- SPARSE SOLUTION MODULES -----------+         +------------- MAXIMUM DISK USAGE -------------+
      HIWATER               SUB_DMAP        DMAP               HIWATER               SUB_DMAP        DMAP
      (WORDS)   DAY_TIME      NAME         MODULE                 (MB)   DAY_TIME      NAME         MODULE
   1024163852   09:32:33    SOL2      368  SOLVIT             3292.875   09:34:35    SOL2      600  EXIT    


 *** DATABASE USAGE STATISTICS ***
 
 +------------------ LOGICAL DBSETS ------------------+   +------------------------- DBSET FILES -------------------------+
  DBSET      ALLOCATED   BLOCKSIZE        USED    USED     FILE       ALLOCATED     HIWATER       HIWATER  I/O TRANSFERRED
              (BLOCKS)     (WORDS)    (BLOCKS)       %                 (BLOCKS)    (BLOCKS)          (MB)             (GB)
 
  MASTER          5000       65536          48    0.96     MASTER          5000          48        24.000            0.511
  DBALL        2000000       65536        6496    0.32     DBALL        2000000        6496      3248.000            6.295
  OBJSCR          5000        8192         318    6.36     OBJSCR          5000         318        19.875            0.025
  SCRATCH      4005878       65536        2832    0.07    (MEMFILE         5878        5046      2523.000            0.000)
                                                           SCRATCH      2000000           1         0.500            0.000
                                                           SCR300       2000000           1         0.500            0.000
                                                                                                            ==============
                                                                                                     TOTAL:          6.832
 
 *** BUFFER POOL AND SCRATCH 300 USAGE STATISTICS ***
 
 +----------------- BUFFER POOL -----------------+      +-------------------------- SCRATCH 300 --------------------------+ 
   OPTION        BLOCKS      BLOCKS      BLOCKS           OPTION    HIWATER               SUB_DMAP        DMAP     OPN/CLS
  SELECTED     ALLOCATED     REUSED     RELEASED         SELECTED  (BLOCKS)   DAY_TIME      NAME         MODULE    COUNTER
  GINO,EXEC        5876        3839        7403              2           1     9:30:06    PREFACE     0  PREFACE       0

HEX20-RESULTS.png

If I set SMEM=7GB in the NASTRAN EXECUTIVE, rerun the analysis and the solution time is improved because the I/O is greatly reduced, in my system is 4 min, 15 secs.

Take a look o your *.F04 file, if you have 128 GB RAM you will be able to increase SMEM to say 20 GB to be sufficient to contain the entire SCRATCH dbset, OK?.

Best regards,
Blas.

 

Blas Molero Hidalgo, Ingeniero Industrial, Director
IBERISA • 48004 BILBAO (SPAIN)
WEB: http://www.iberisa.com
Blog Femap-NX Nastran: http://iberisa.wordpress.com/

Re: Slow NX performance

Phenom
Phenom

Hyperthreading should always been off for Nastran, Abaqus, etc...  At least with Intel processors.  It comes from the throttling Intel does.  That tendency isn't necessarly true with AMD Ryzen, we will post some data soon.

 

In general though, I wouldn't bother with SMP, especially with such a small problem: it takes 1:31 to solve on 1 processor and 1:27 with SMP=4 with NXN 12.0.1  Smiley LOL

 

On a side note, the same model with MSC Nastran 2018.1 takes 0:47 on one processor and 0:43 with SMP=4

 

Re: Slow NX performance

Pioneer
Pioneer

Tentech,

 

Are you running the original model with parabolic elements and 948,321 nodes, or are you running with linear elements and 241,211 nodes? The model with linear elements and 241,211is a smaller model and it makes sense to me that it would not show a large benefit from SMP, however, the larger model made from parabolic elements and 948,321 nodes did show sensitivity to SMP resulting in a 19.2% improvement in solutions time. This the case for the particular workstation I was using for these trials. 

 

I guess I'm still a little disheartened, as to the solving efficiency of NxNastran when compared to ANSYS.