High Everybody,

There is a recurrent discussion on the limit for people from industries to carry up-to-date simulations with their commercial codes. For instance, LES is still out of reach because it takes to much time as industrial solvers do not scale very well in parallel. Has anybody an idea of the limit of current commercial solvers for parallel calculations: maximum number of processors before the efficiency drops, for instance?

We currently have applications where we use 32 workstations in parallel. We are working with our vendor to port the code to 100 workstations.

Tom

Dear jojo,

The ANSYS CFX solver is known run for a large number of processors. In fact, version 10.0 was limited to 512 processors, and version 11.0 has been upgraded to 1024 processors.

Some users has reported successful calculations with more than 512 processors using the pre-release version 11.0.

The maximum number of processors before the efficiency drops has a "depend" answer. The efficiency is tied to the size of the mesh. There is an overhead while communicating data among processors (increases with number of processors), if you start to increase the number of processors for a fixed mesh size, the cost of the calculation (CFD part) diminishes per processor and only the overhead is left (which does not scale)..

Hope that helps,

Opaque

CD-adapco has a couple of pages of benchmarks on their web site. They have timing results listed for a bunch of models run over a good cross section of available hardware. Although code specific, this might help answer your question.

http://www.cd-adapco.com/products/ST...320/index.html

AcuSolve is also very scalable - also with LES or DES.

Thanks a lot everybody