Hi all,

I want to simulate the flow around a UAV in Fluent. My grid contain 5 million cells. Can any body tell me what is the require computer resource for this work? Do I need parallel processing, certainly? Do I need to use Linux or Unix as OS? Do I need to use Fluent in batch mode instead of GUI? OR NOT?

thanks in advance,

Max

I think you need computer to parallel processing. Parallel computing model of boiler (1.5m cells) on 6 processors with 2.6 GHz takes me about 14 hours for example.

But I have solved a problem with 1.5m cells on a PC of 3 GHz and 2 GB RAM during something around 10 hours.

flow with combustion in 3D ?

oh no, Sorry, just turbulent flow. But what about my present problem?

You do not need parallel processing, certainly. Computing on 1 process will be only longer, but cheaper. Licence on 1 process is cheaper then on 6 for example processes.

You do not need parallel processing, certainly. Computing on 1 process will be only longer, but cheaper. Licence on 1 process is cheaper then on 6 for example processes. But if computing on 1 process will be long and you will to have make many modyfications in your model, computing in parallel version will be better solution.

Is there any way to understand whether I need the parallel processing or not (a final answer)? there is no problem regarding preparing license for fluent.

your problem will be the memory For a 5 millions cells mesh, you need about 5 GB Ram. Parallel processing will "just" speedup your calculation. Now if you have an architecture which support 5GB Ram, you won't have problem. But if for instance, if you are under windows XP 32bit, then you are in trouble, thus XP 32-bit will not support up to 3GB Ram (it's my experience, but I am not IT-specialist)

So which OS will I need? and how much time does it take to solve my problem without parallel processing?

This depends on whether or not you can use the segregated solver, which needs less memory than the coupled solver. It helps even more if you can use an incompressible approach. You will need a 64-bit version of Fluent and the operating system, and in theory it will then be possible to get away without parallel processing. In practice, that will probably be too slow to give you much useful data. A minimum spec for this problem will be to go with a quad core single socket computer, 8 GB of RAM and a license for 4 parallel processes. The parallel scaling over 4 cores is not brilliant, but you will still get a benefit from it. If the budget will stretch that far, rather get a dual socket computer, with dual-core or quad-core CPU's. Remember that computers are much cheaper than licences, and parallel licenses are not that expensive compared to man-hours.

If we consider that I have to solve a compressible problem and for at least 40 times, do I still use a PC?

Compressible, 5 million cells, 4-way parallel, I guess that you will need about 40 hours per data point on 3 GHz quad-core (this is a really rough estimate ...) so estimate from that. Basically, it looks like a bit more than a 2 month job. I reckon it makes sense to go for a dual-socket workstation, but the call is marginal.

Could you explain me more about dual-socket computer or workstation (as you mentioned)? I don't know anything about it.

Thanks a lot

OK. There are many different options for CFD computers, but in general X86 (i.e. PC type) processors are now by far the best value for money. At the bottom end of the scale, a good CFD workstation is basically a high end PC, typically using a single quad-core processor from Intel or AMD. When it comes to number crunching, there is not too much to choose between the two brands. However, a big limitation of this style of computer is that they typically have only 4 memory slots, and can support a maximum of only 8 GB of RAM. The assumption here is that only 64-bit operating systems will be used.

At the other extreme, what you really want is a cluster, typically using rack-mounted or blade elements.

If you want something better than a PC, but can't afford a big cluster, a very good option is a dual processor workstation, as sold by a variety of vendors (Dell, HP, IBM, BOXX and Appro all have good products, typically with very good 3D graphics cards). When equipped with quad-core CPU's, this will give you effectively 8-way parallel, if you can afford the parallel licences, without the hassle of a cluster. Some of these motherboards have up to 16 memory slots, so you can get 32 GB of RAM installed. Bear in mind that you won't get perfect parallel scaling, so if you are on a tight budget, dual core CPU's are likely to be more cost effective. A dual-socket shared memory computer like this is more expensive than two individual "distributed memory" PC's, and is unlikely to be any quicker than the 2 PC cluster. However, it is far more convenient, and generally much more reliable.

Beyond this, you do get 4-socket shared memory workstations, but they are quite rare. 4-socket server nodes (i.e. rack mount, noisy and with rudimentary graphics processing capability only) are a bit easier to find. Once again, expect to pay more than you would for 4 PC's, or 2 dual socket workstations. For computers built with 4-socket motherboards, AMD have a big performance advantage over Intel.

Thank you Charles for your beneficial answers. I need to consult with some other on your answer and I will report them to you tomorrow.

Best Regards,

Max