The general purpose code like CFX needs a couple of days to run my case with relatively simple geometry. But I learnt that some application focused code just needs only 30 minutes or so with a geometry looked much more complicated than mine. Just curious what the trick is? Thanks.

could you perhaps share with us what your case is and what that "some application focused code" is?

Here is the link for the provider of the program, http://www.futurefacilities.com/ Their CFD code is specially designed for data center cooling. I've seen the demo, the geometry is rather complex. I am really very surprised that it takes just that little time to run. I've tried to model a rectangular shaped plenum with CFX. The thing is that its height is very small compared with the other two dimensions. So the mesh size is pretty large.

There are a lot of layers to this particular onion. But generally, yes, specialised codes can be a lot better for a given application than the general purpose ones. Very often the specialised codes can make big assumptions about which aspects of the physics need to be modelled, and which can be dropped, or efficiently approximated in some other way. It is left to the user to decide if these assumptions are appropriate or not.

But there is a general principle at work here ... If you are using these specialised codes for design, you are not trying to model the real thing accurately. Rather, you are using it to make decisions that will enable you to make a good design. The distinction is very important, and it relies a lot on your knowledge of what you are designing.

So what you need to do is to get hold of the specialised code, and try it on something that a) you know the answer for, and b) is very similar to what you are designing. If you don't have access to such data, you need to do more research before relying on a specialised design tool that you don't know well enough.

one thing you have to realize is that there is no magic in CFD, all the flows are described by NS equations and those are PDEs and you discretize them and end up with a bunch of algebraic equations and yes you could get into this big discussion on explicit/implicit time and coupled/un-coupled solver etc... but on a large scale of things datacenters from a CFD stand-point present no challenges, it's just an airflow with buoyant effects around a bunch of boxes... in other words "task specific" in this case could only refer to some lovely graphics, since there is no complex physics involved as you might have for example in acoustics or combustion. If you are working for a manufacturer, ask the guys at CFX and the guys at the company you mention to run a "blind" test and compare the results to your experimental data and if there is a sufficient number of licenses involved I am sure any "major" code vendor would be happy to oblige and provide more accurate results, truly converged on a professionally build mesh and in no-time... it is a nice animation though, but that's all there is... and you do not need CFD to do that