Learning Forum

[Talla]

Hi guys,

I am using a big model which has more than 900000 elements. So solver goes automatically from in core solution to out core solution and this will cause very large time calculation( like 14h for modal analysis).

i have notice the amount of memory required for in core solution is 56985.9 MB and i have 64G for the RAM memory. so i don't understand why its solving with and out of core. actually i don't understant how the allocated memory is estimated, mine is very low.( please see the picture bellow)

does anyone can help to fix this please?

thank you!!

 

[peteroznewman]

Hi Talla,

The difference between 57 GB and 64 GB is only 6 GB. Windows OS is going to consume 4 GB, then you have other programs open, like Workbench. There is only a 2 GB margin left, and the in-core memory estimate is approximate.

Sometimes, even if you have a bigger margin, if you have been using the computer for a while, the memory can get fragmented and there is no single block of free memory as large as it needs to allocate, so it uses out-of-core.  This can be remedied by restarting the computer and trying again.

There is a command snippet to force the solver to try in-core.  If you are using Distributed solver it is:

DSPOPTION,,INCORE

Note the two commas, that is not a typo.  That will force the solver to start incore. However, if it is a multistep solve, it may switch to out-of-core at a later step if it needs more memory than is available.

[Talla]

thank you Peter for your response, i undertand what you explain and i will try what you suggest. i have also notice one other thing that seems to me very weird. the computaion takes less time when using  a lower number of core than a higer. for exemple that takes 12h for a 16 cores and 6h40 for a 2 cores. i had try to understand why a i have this and finally i think that when the process is dirtriuted, the sum of memories requested by each core will be higher than when we have less core. is that corect for you?

[peteroznewman]

That doesn't sound right.

If you have a single computer with 16 cores and a model that solves in-core, try solving on 15 cores. Then rerun it on 2 cores, it will take longer.

If you have run out of RAM and are solving out-of-core, then it might be the case were more cores slows down the solution relative to 2 cores.

[Talla]

ok thank You peter, i have another question but i think that it will be better if i put it in another post.

thanks for your time