Improving simulation time

[mark1987]

Hi all
I am looking to upgrade my PC to increase the speed I can process CFD models. I currently have an old 32 bit dual core Intel Xenon with 4 GB RAM. I am unsure as to what aspect of a new machine will give me the best improvement – i.e. processor speed or amount of RAM.
A possible machine I am looking at is a new quad core or 64 bit Dual Six Core model with 32-128 GB RAM. Can anyone advise on how much faster this new machine would perform when compared with the old machine, or advise on a better option? Would more RAM speed the simulation or would that just only limit the number of cells that can be simulated and not make any appreciable difference to the simulation time?
A typical model would be 4 meshes with a total of 2 million cells, although using more meshes would be possible to make better use of multi-core processors.
Any response would be much appreciated.
Mark

 

[IRstuff]

Why don't you ask your CFD software supplier? They should know best what will help their software.

TTFN
faq731-376

 

[braddles90]

Increase the RAM to allow you to use/store bigger meshes, and upgrade the CPU to increase solving speed.

I don't know off the top of my head the numbers, but ANSYS has some guidelines as to how big the model has to be before you see any real decreases in solving time when using parallel processors. Have a look around for ANSYS parallel processing guidelines, if you start looking at a dual CPU solution (or a cluster, if you're well funded).
An SSD could also be a good upgrade, if you run simulations that write data to the hard drive a lot (i.e. saving transient results during a simulation). Decreasing the access/write times could give a noticeable speed boost in-between calculation steps.
Be careful around multiple core CPU's as well - most software (well, CFX at least) only gives licensing up to 4 cores on a CPU before you need to buy more/ a different type of licence (so hyperthreading, or 6-8core CPU's mean you potentially have a bunch of CPU you can't use). I'd suggest going for the fastest clock speed 4-core CPU you can get, maybe 2 if your meshes are big enough (sounds like they are, once they're in the millions).
As IRstuff mentioned, though, it's more than likely highly software specific - give a call to the tech help desk, they're most likely more than happy to help.