How to speed up the solution

[JordanCN]

Sorry about the lack of details on this one, but here it goes. I have a user that had a 1.4 Ghz PC with 768 MB of RAM. They wanted to speed up some of the models that take about 2 hours to run so we updated them to a 3.2 GHz computer with 2 GB of RAM. I thought this would really speed things up, but they say that it has not.

I have see posts about AMK memory or some variables that need to be setup sometimes to get the computer to use all memory. Is there a list of tips that I can review to help speed up the solutions?

 

[Stringmaker]

Really need to know more of what the scope of the analysis is here to give concrete answers. Increasing the size of the workspace will be of negligible help if ansys was only utilizing a portion of the memory available when using 768 MB. I'd be very surprised if the user didn't have a noticable speedup with the faster processor and more memory. It may not be a scalable factor based on processor speed and amount of memory but there should be a good bit of difference in what you described if all of it is being utilized. I'm not sure which environment you're working with but I believe the AMKMEMORY parameter is only valid within Workbench. Windows task manager is especially helpful in determining cpu and memory utilization to find where the bottleneck may be. Here is a list of things that may help additionally:

-Choose a different solver. If you have an analysis that is of small to medium size and can be ran completely incore (ie. using only physical memory) the sparse solver should be the fastest if it's applicable for that analysis type.

-If possible utilize a machine with significantly more horsepower for the analysis.

-Create a coarser model if possible. This may not perhaps be the best approach in this case but I had to throw it out.

-Use lower order elements. For instance 8 node bricks instead of 20 node bricks. Avoid 4 node tets though at all costs.

-Perhaps try building an entirely or atleast partially hand built mesh. This will give a significant decrease in node numbers and great increase in element quality versus those typically obtained with an automesher. This will take some time though and user vs. computational time should be weighed here.

-Run analyses not during regular business hours (ie. overnight).

If this becomes a recurring issue serious thought should be given to purchasing a designated machine of an appropriate capability to solve your models. If someone is rendered useless for two hours daily because their computer is busy solving a model it doesn't take long before spending several thousand dollars on a more efficient or even additional computing resource can easily be justified.

 

[Mech151]

Dell just came out with a new line of PC's. 16GB ram, 2x dual core processors for very economical (490 precision). You would be lucky to get 1gb ram available for ansys on a 2gb ram machine (thanks Mr. Gates). Also, RAM has come down, you could consider jumping to windows 64 and adding some additional ram to your computer.

Are you using ansys or workbench? If you are using workbench you can add a solution information branch, and it will state the in-core memory requirements. It would be good if you would post things like: ANSYS or Workbench, how many elements, how many nodes, what solver, etc.

Also, to free up a little more ram, go to c:\windows\Prefetch and delete everything in the folder. These are items that windows holds in memory from the time you boot up to facilitate quick launch. Deleting items from here will not impact functionality of anything.
Regards.