Will Nastran NX use more than 4 processors? 1

[jerzy]

Hello forum members,

I'm considering buying a new workstation. I use FEMAP and NX Nastran and often run non-linear Solution 106
problems. I was looking at the Dell 7600 with 8-core processors. (I currently use 4-cores.)
Will NX Nastran use the 8 cores? I would probably get 64 Gigs of RAM. Would there be any performance improvement
if I got 15K RPM SAS hard drives?

Thanks

Jerzy

 

[BlasMolero]

Dear Jerzy,
Of course!, this is native of NX NASTRAN, simply include the nastran keyword PARALLEL=value, where "value" specifies the maximum number of CPUs selected for shared-memory parallel (SMP) processing in several numeric modules. SMP processing reduces elapsed time at the expense of increased CPU time. The default is 0, which specifies no SMP processing. If “parallel=1", the parallel algorithms are used on one processor.

Note: If you need to vary the number of SMP CPUs during a job, you must set either the “parallel” keyword or SYSTEM(107) on a NASTRAN statement to the maximum number of SMP CPUs that will be requested. Some systems cannot process a DMAP request for CPUs in excess of this initial value.

Example: nxnr nastran example parallel=2
The job is run in SMP mode on a maximum of two CPUs.

Regarding RAM memory, 64 GB is perfect, go for it. And if pricing is available, mounting super fast SSD of say 500 GB is critical, you will experience a brutal response, great!.

Best regards,
Blas.


~~~~~~~~~~~~~~~~~~~~~~
Blas Molero Hidalgo
Ingeniero Industrial
Director

IBERISA
48011 BILBAO (SPAIN)
WEB:

http://www.iberisa.com

Blog de FEMAP & NX Nastran:

http://iberisa.wordpress.com/

 

[jerzy]

Hello Blas,

Thank you for the quick and helpful reply.

I also want to thank you for all the information you have supplied in this forum. I have learned much from you and always look forward to studying your replies.

Best regards

Jerzy

 

[loki3000]

a nvidia tesla K6000 card would also speed up the calculations.

 

[nlgyro]

The awful truth is that SOL106 is not scalable on NASTRAN, unlike linear static or dynamic solution sequences that are scalable till 8cpu's

What could help you is tonnes of RAM, but again nastran is limited to 8Gb of RAM on windows.

You could try a linux platform where the 8Gb RAM limitation is not present, or try running your jobs on scalable implicit non-linear solution sequences, which NX-NASTRAN offers.

Good Luck!

 

[BlasMolero]

Hello!
This is not exactly correct, NX NASTRAN provide two executables in Windows x64 platform: regular NX NASTRAN LP-64 executable that can allocate up to 8 Gb of RAM memory (this is the solver I run by default when the job do not require more than 8 GB RAM, allowing most of the RAM for I/O OS cache), and NX NASTRAN ILP-64 executable that can allocate approximately 20 million terabytes. Practically speaking, there are no machines currently supporting more than half a terabyte, thus the amount of memory these executables can allocate is only limited by the amount of memory installed on the machine.

When the LP-64 executable is used, the bytes_per_word is 4. When the ILP-64 executable is used, the bytes_per_word is 8. This difference is important when you are specifying memory in "words" with the “memory” keyword.

Also, regarding Parallelism in NX Nastran, we can say the following:

NX Nastran supports both shared memory parallel (SMP) and distributed memory parallel (DMP) processing:

SMP is used only for lower level operations such as matrix decomposition and matrix multiplication for all solution sequences, including NonLinear Static analysis (SOL106). Therefore, as long as suitable hardware is available, all solutions can utilize SMP processing.

The DMP is based on domain decomposition on geometry domain or frequency domain, or load domain. DMP methods achieve their solution speed by dividing the FE model into smaller pieces to be solved simultaneously. This division is performed with respect to geometry or frequency range individually or both at the same time. Although each processor is working on its own partition of the geometry or frequency
range, it communicates with the others to share information. Once the solution is complete, the results are merged, creating a single result file.
They are for linear static analysis, normal modes analysis, and frequency response analysis, respectively. Modal frequency and transient responses require computing modes for modal space. Therefore, DMODES can be applied during mode computation. In design optimization, if it involves computing modes, DMODES also can be activated.

In summary, DMP computational methods support the following solution methods in NX NASTRAN:
• SOL 101 Linear statics.
• SOL 103 Normal modes.
• SOL 105 Buckling.
• SOL 108 Direct frequency response.
• SOL 111 Modal frequency response.
• SOL 112 Modal transient response.

Best regards,
Blas.

~~~~~~~~~~~~~~~~~~~~~~
Blas Molero Hidalgo
Ingeniero Industrial
Director

IBERISA
48011 BILBAO (SPAIN)
WEB:

http://www.iberisa.com

Blog de FEMAP & NX Nastran:

http://iberisa.wordpress.com/

 

[nlgyro]

The core non-linear iteration module, NLITER is not scalable on multiple CPU's using SOL106, and this matters when one talks about scalability on multiple CPU's

The other option is giving huge amounts of RAM so that the module can run in-core, using the ILP64 binaries of the solver, but what happens to users use NX6.0 (for legacy reasons) where ILP64 version of the solver was not available on windows?