I will be getting a new PC for my calculation, which is a pretty complicated combustion chamber model with way over one milion cells probably ( the grid is not yet finished). Since I am suppose to decide on the hardware, here my questions. Is Opteron much faster than Xeon when we compare similar price range? Does generally Intel give You fewer problems with the hardware? I heard that Fluent 6.0 and 6.1 don't work very well on Opterons. Is it possible to connect two PC's one an AMD Opteron and the other Intel Xeon to do the calculation? I will definitely be using Windows OS, probably 2000, maybe XP but that is unlikely. I will be grateful for any remarks since I am new to CFD (this is my diploma thesis) and also do not know that much abiut hardware as to answer above questions. Thank You.

There shouldn't be a problem mixing Opteron and Xeon machines. I have access to a Dell Precision 670 with two Xeon 3.6 GHz CPU's and a Fujitsu Siemens Celsius V810 dual Opteron 252 system. Running 32-bit Windows executables, the Opteron appears to be about 10~12% faster. If you believe SPEC figures, with properly optimised executables for each platform the difference can be up to nearly 35% in favour of the Opteron. Sounds like you need a lot of memory, most dual CPU Opteron setups have eight memeory slots, whereas the Xeon boards generally only have 6. That is academic if you're not running a 64-bit system though.

Over a million cells with Windows is a recipe for disaster. Windows memory management has some issues, and generally won't allocate (MALLOC) more than 2GB for any process. A million cells is probably a bit over this limit. You may need to go 64-bit.

I heard that windows gives up to 3.3 gigs for a single process. Going 64-bit is highly unlikely due to the other software limitations in the laboratory. Using Windows is the only option for the time being. I also am worried about the cell count. This is my first simulation and a complicated one to start with. I hope 3 gb of memory will do the trick though.

One more thing. A friend is doing a compressor analysis on a dual 2.0 Xeon machine with 2gb memory running Win2000. He has over million cells and manages to perform the calculation, albeit only with segregated solver.

That sounds about right. Different solvers have different requirements. Multi-block structured solvers need minimal memory, segregated Fluent is a moderate requirement (bit more than 1 GB for 1 million cells, but it depends on which other models (such as turbulence) you have switched on), coupled Fluent needs a bit more. 32-bit Windows on a computer with 4GB of RAM will effectively give you about 3.5 GB that you can use, but each process is limited to 2 GB max, so you really want a dual CPU machine, and run two CPU's in parallel. You should be able to run up to two, maybe 2.5 million cells, but you are going to wait a while!

Thanks a lot for Your response. One thing I am wondered about - how can I benefit from dual processor setup when it comes to max memory allocation since I still will be running one instance of Fluent? Do you mean that it is possible to launch 2 instances of Fluent on dual PC setup and divide one calculation among them? As for turbulence models, I am still deciding whether or not use the wall function instead as near wall turbulence is of no big importance in my problem, except for the air blast piece maybe.

I think you should run Fluent parallel.

Dear Jagus:

I have used fluent 6.0.12 and 6.2 with AMD sempron and Duron processors for about 3 years. I have not seen any problem with AMD processors and FLuent. fluent wngin is written in c++ language. on of the most advantage of c++ is that works on a wide range of work stations and processors without any problem. so be sure, no problem accures with AMD. Regards