I work with StarCD on a PC that has the following characteristics:

CPU Pentium III 450 Mhz

RAM 128 Mb

If I want to increase it's performances in terms of CFD runs, should I change CPU or would it be better to add more RAM Mb??

Thanks a lot, Fabrizio

(1). Use a faster CPU will always increase the performance. (2). Add more memory will in general allow you to run a bigger job faster.

I have plenty of RAM and processor, but the Xserver/graphics slow down the user interface significantly. My current grid is ~1.2 millions cells and it is extremely slow to move the mesh for viewing/applying boundary conditions. Is there a faster Xserver than Reflection X, or a better video card to run? I'm running STAR on a WIN NT4 PIII 733X2 machine.

Thanks

Go for more ram, I run star-cd for nt on an old pentium pro that had 128 MB of ram. I upgraded as much memory as I could (couldn't get any more simms) to 224MB. Much more stable operation, less paging hence faster. If you can put in as much ram as possible.

allan

Prostar uses hardware acceleration only in the "term ,,, extended" mode of the glm version. Otherwise it is just processor and bus speed which helps. So you can use a fast Unix workstation to speed up your work.

which videocard do you run?

If you are already paging or swapping, no matter how much faster your cpu is, you won't get any additional speed, because your bottleneck will be disk to memory transfer time. If your problems fit in the physical memory that you have, then buying a faster cpu may make your runs faster. However, even if you fit in memory, depending on the caching characteristics of your system and the problem you are running, you still may not see much of a performance benefit. If you are running jobs greater than 200000 cells, you need more memory first.

(1). Yes, this is very important. Since the hard disk I/O operation is much much slower than the speed of CPU, one can easily detect the slow down when running a big job with small RAM. (2). 125Meg is relatively small for 3-D problem, I think, if you are solving 3-D problem, you need at least 512Meg of RAM, which in general should be enough for 350K cell size. (3). So, the first step is to make sure that you have enough RAM for your problem. After that, the speed up will be proportional to the CPU speed. (4). Three years ago, when I was running 3-D problem using a commercial code, my colleague couldn't even post-process the data file I created, because my data file exceed his workstation's RAM capacity of 125Meg. In today's standard, 512Meg is a minimum for 3-D problems.

Your numbers are completely wrong. Plaese read the manual. This is not the Fluent group

A Fluent execuatble (segregated solver), is roughly twice as large as a STAR executable for a given mesh density and problem definition.

Ron

(1). Yes, you are right. Those numbers are not related to the particular code you are using. (2). I only tried to use it as an example. Even with Fluent, those numbers are not right at all. (3). No one should trust any number mentioned here. The message was: the RAM memory should be large enough to handle the particular problem at hand (that size changes from problem to problem even if you have the same number of cells). And this seems to be a common sense for any CFD user. (4). I think, 512Meg should be able to handle a typical problem of 350k cells, because normally on the same workstation, you have other processes running. So, a 512Meg RAM workstation is not dedicated to run just one job alone, based on my daily experience. (5). As a matter of fact, I am alway running two jobs on the 512Meg workstation, one is the old job and one is the new job. (6). So, I was not saying that one code is less (or more) efficient in momory handling than the other. I was trying to point out that if you have a 512Meg workstation, you should be able to handle a 350k cells job in the typical working condition, where there are always many processes running on the same workstation, and in my case, normally there are two jobs running at the same time. (7). The size of the compiled program is really not the issue, because, the user is interested in the solution accuracy rather than the cost saving in RAM. The price of RAM is really very low right now. (of order $1.0/1Meg RAM) (8). I am sorry that the numbers I used was interpreted as the optimum memory requirement of the code you are using. It merely used to serve as an example in my working condition.

I have some experience of running StarCD at different PC's. Yes, the memory is more important than CPU speed. But there is also another way - you can use Linux version of StarCD, this can decrease calculation time nearly twice, if running from console with no X server started.