Which server to run Abaqus for fatigue testing (elastoviscoplasticity)

[RGX124]

Hi,
We are considering to buy a faster server for running Abaqus for multiple users (4-5).
We have already a Dell Poweredge 1950 16Gb Ram - 2 quadcore Xeon 5355 2.66GHz- 2 SAS disks 73G (15krpm).
Since other applications are already running on the PE1950, we would like to set up a cluster with the PE1950 as a front-end server and another server as the calculating server.
I would like to know what characteristics are important for this server: RAM? Disk's speed? Disks capacity? Processors? Number of processors?
Is it faster to have a server with 2 processors or 2 servers with one processors?
The server we would like to buy is:
Dell PE 2950 III
32G Ram - 2 quadcore Xeon 5460- 2 SAS disks 73G (15krpm)

Do you have any ideas of what are THE IMPORTANT CHARACTERISTICS for a server in order ro run elastoviscoplasticity simulations of fatigue loads as fast as possible?

Thanks for you help.

 

[xerf]

Some points of view:

1. The amount of RAM is a critical aspect if you run large models. However, having 32Gb seems good.

2. If by fatigue, you understand structures under cyclic loads where you simulate the entire loading path cycle by cycle then you should be prepared to manage a large amount of data on HDD. I had simulations where each case produced 60-100Gb of .odb files (even though only a few frames per cycle were stored). In this case, I would say that the disk capacity is more important than the disk speed especially if you need to transfer large amounts of files on a regular basis in order to free space on the HDD contaning the ABAQUS working directory.

3. What would be the main task for the front-end server ? (I would not "waste" a server for creating models in CAE unless you have to mesh parts with a complicated geometry.)

4. Regarding 2 x 1 CPU machine Vs. 1 x 2 CPU machine, I do not know if I can provide the right advise. However, for some reason in years of using Abaqus on various machines I have not noticed a significant improvement on running a single job on a 2 CPU machine vs. running the same job on a 1 CPU machine. Sometimes I have had the feeling that for running a single job the CPU speed is more important than the number of CPUS. The computational speed will decrease with the number of jobs run in parallel but will be more noticeable for the 1 CPU machine.

Best.

 

[gurmeet2003]

Running a job on a 2 cpu machine is much faster than running the same job on 1 cpu. I have a 2 cpu machine and I can choose to use 1 or 2 cpus. When due to an omission 1 cpu was picked, the job took much longer. This becomes apparent if you compare the cpu time with the clock time. Therefore I am surprised by Xref's comments.

As regards RGX124's question about one 2 cpu machine vs. two 1 cpu machines, I am not clear about the intent of the question. Do you really mean 1 cpu machines or the numbers were just picked for illustration. If you are really talking about two 1 cpu machines I would go for one 2 cpu machine.

Gurmeet