This content has been marked as final. Show 3 replies
From testing I did, you typically get a 30-50% solve time improvement with 4 or 8 CPU's, especially on larger models. I'm talking in the 200,000 up cell count. I did models that solved in a matter of minutes to hours.
On smaller models, less than say 20,000 cells, there is marginal value of multiple CPU's.
The benefit seems to consistent across simulation types too, so it's not restricted to just steady state or thermal.
What's really cool in 2009 with the batch processor is that you can specify how many CPU's each job is permitted to use, so if you know it's a small model and you have a large one, you can tailor the CPU count to suit.
thanks for your response. Has anyone else had any experiances with 2009?
If you loose a face for some reason on a BC ther eis lack of robustness in the rebuild function to get the fix to stick. Some time it requires going back to SWX and others it required an exit and reload to get he rebuild to execute properly. Not a big deal but a bit of a pain. Otherswise it seems to work as well as before.
The scaling for hte multi CPU's is a long way from ideal but hopefully they can improve that over time. It is definitely a really welcome capabaility even though I am a touch greedy ont he subject. Many CFD codes get near linear scaling on low numbers of CPU's. Hopefully Flow Sim. can get near that as well.
The call outs for BC's are very handy as well. It would be nice though it they enabled it for all items in the specification tree - like goals, material definitions, disabled components, inital conditions, mesh controls, etc. It would also be really nice if for volume specifc items the callouts pointed to the volumes C of G and for surfaces it pointed to the face centroid, and the centers of anything else that gets added. Further on call outs I have had cases where multiple call out leaders are applied to a single body with uniform heat genration - a bit confusing as one might think that mutiple powers or some other distribution than uniform is implied. It would also be nice if the renamed lables in the tree were passed to the call outs - it would remove a lot of writing from reports to ensure clarity.
Further coloumb heat generation would be handy. Say you pick a volume heat source and give it resistive properties in given directions, supply a current exit location, volume unifrom "current generation" and then the heat genration would be proportional to the path length and current density. admittedly, I haven't thought that one out that well but something that gets it in theball park would be really handy when dealing with battery cooling which is becoming a very hot topic these days. Punn fully intended.