CFD Online Discussion Forums

CFD Online Discussion Forums (https://www.cfd-online.com/Forums/)
-   FLUENT (https://www.cfd-online.com/Forums/fluent/)
-   -   64-bit FLUENT (https://www.cfd-online.com/Forums/fluent/36860-64-bit-fluent.html)

Aaron June 6, 2005 13:05

64-bit FLUENT
 
Does anyone have experience running FLUENT 6.2 on SUSE or REDHAT (64-bit versions) with AMD64 or EMT64 processors? I've heard little "bits" of info through the grapevine that engineers/researchers are seeing increases from 2GB to only 4GB of memory (per process) when going from the 32 to 64-bit architectures, operating systems, and FLUENT 6.2. Is anyone able to access (actually observed, not theoretical...) more than 4GBs on 64-bit systems?

Thanks, Aaron

Aaron (again) June 7, 2005 14:31

Re: 64-bit FLUENT
 
Since my first post, I did some more research into this question and found that the answer is likely, "NO". That is, even in a 64-bit operating system, 64-bit architecture, and running version 6.2, FLUENT can not take advantage of more than 4 GB of memory per process.

The previous limit for 32-bit systems was 2 GB (sometimes 3, if the right settings were switched on in Windows), which corresponds in my experience to a maximum of about 4 million cells in a hybrid grid. For big multi-processor computers (SMPs), that wouldn't matter during the calculations, but as others have reported in posts elsewhere on this site, the problem is in pre- and post-processing where the entire grid and the information calcuated at each of the cells is read into memory on a single processor. So unless the solution is pre- and post-processed in parallel (is this possible?) or in pieces, then 4 GBs appears to be the ceiling...

Fluent, Inc. reports results of a benchmark done at 9 million cells for turbulent flow through a transition (tapered) duct. The mesh cells in that case were hexahedral and I believe this is why they were able to get so many. Also the geometry is rather simple, so perhaps the entire model in that case will fit into something smaller than 4 GBs and maybe even smaller than 2 GBs.

So... as far as deciding whether to spend the money to upgrade to a 64-bit computer for running larger CFD simulations, particularly when the decrease in wall-clock time is only 1/3 of total, then I just don't see that its necessary. Increasing the number of processors and maxing out memory per node (2GBs per processor) might be useful, but getting a whole new 64-bit system just doesn't seem like a good way to spend money.

Anyone wish to comment on that? Anyone think that it is a good idea to upgrade to 64-bit? Why or why not?

Also, could someone explain (perhaps someone who has seen the FLUENT source code or knows about the hardware) why the software can not take advantage of more memory than it does, particulary for a 64-bit system that has a theoretical memory limitation of 16 terabytes per process? Does it have to do with the network interconnects? Bus communication? Some other reason?

Please feel free to share your thoughts, knowledge, etc. This seems to be a relatively unknown area so I would personally appreciate any feedback.

Thanks! Aaron

zxaar June 8, 2005 22:35

Re: 64-bit FLUENT
 
i have four 64-bit linux machines and yes the maximum memory that i am able to use is 2.9gb, though my rams are 4 gb, not very much sure why?

further after switching from windows to linux i did not see much speed up comnpared to windows , though lot of people seems to say that linux is far better than windows when it comes to speed. i am no fan of windows but din't really see much advantage using linux.


All times are GMT -4. The time now is 16:05.