differences between FLUENT 32-bit and 64-bit executables
|
It is well-known that the FLUENT 64-bit executable, which can be started by adding -64 the FLUENT command line, e.g. %> fluent 3d -64, is better than the 32-bit executable for problems in serial that require large amounts of memory - see for instance Solutions 232 and 344. Are there any other known performance differences between the 32-bit and 64-bit executables? The 32/64 bit refers to the size of the addressing used in the executable. The size of a pointer in the 64 bit executable is twice that of the 32 bit executable, which allows you to access more memory space. The size of the reals (floats) and ints are the same in either executable (storage used by the data itself is the same size). So in going to a 64 bit executable, I believe you are just doubling the size of storage required for the pointers in the code. FLUENT does use a lot of pointers, but I do not have exact numbers for you. Also, this will be affected (to a certain degree) by what models are enabled (less so than total memory useage though). I don't have a rule of thumb for you but a spot check on a 170K hex mesh turbulent flow with heat transfer used 22% more memory in 64 bit than in 32 bit on an sgi machine. There should be no difference in execution speed between 32 and 64 bit. The only caution here is that this assumes that the platform's 32 bit and 64 bit math and other system libraries should perform the same (usually true). There should be no difference in parallel performance, however you may not have the same mpi options since some vendor or third party MPI libraries may not be ported to 64 bit mode. |
||
|