- 积分
- 0
- 注册时间
- 2010-10-11
- 仿真币
-
- 最后登录
- 1970-1-1
|
发表于 2010-10-11 22:22:41
|
显示全部楼层
来自 上海杨浦区
HyperXtrude是支持并行计算的
目前的并行方式是SMP,但效率跟平台相关
这一点在在线帮助上已经写明了。
(顺带提一句,听说11.0中会在计算效率上有大改进,敬请期待吧)
Navigation: HyperXtrude > User's Guide > Guidelines for Simulation >
Running Jobs in Parallel
When running multiple HyperXtrude jobs in parallel, keep the following items in mind:
• It is achieved using environment variables.
• Environment variables set to activate multiple processors vary from system to system.
• The following are for Linux (example shows in csh setting it to 2)
o setenv MKL_NPROCS 2
o setenv OMP_NUM_THREADS 2
• For other systems, one of the items below will work - so setting all of them is not a bad idea.
o OMP_NUM_THREADS
o MKL_NPROCS
o MP_SET_NUMTHREADS
o P_NUMBER_OF_THREADS
o MLIB_NUMBER_OF_THREADS
• Please be aware that this may not improve the speed at all, and on some cheap motherboards with fast CPU, it may actually run slower.
• We observed 20% speed-up at the best for OptiStruct with large in-core solutions.
• However, HyperXtrude always uses out-of-core solution (as the problems are large) and hence it is I/O intensive. Hence, the gains will be less.
• It is possible to force HyperXtrude to use in-core solutions. However, for the kind of problems we solve, you needs over 20 GB Ram. To force this, you should set MemoryForSolver to -1 in the tcl file.
o pset MemoryForSolver -1 |
|