Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.
How to redeuce memory requirement during comsol simulation
Posted 6 dic 2012, 18:49 GMT-5 2 Replies
Please login with a confirmed email address before reporting spam
I am new to comsol. I recently built a simple 3D model in comsol.
Oops - it used way much more memory than I expected. I used to build models with
millions of nodes and never had any issues. But now the simulation would crash whenever
the number of elements got close to 1 million. So to all the gurus out there -
what tricks do you know can reduce memory requirement during comsol simulation?
Oops - it used way much more memory than I expected. I used to build models with
millions of nodes and never had any issues. But now the simulation would crash whenever
the number of elements got close to 1 million. So to all the gurus out there -
what tricks do you know can reduce memory requirement during comsol simulation?
2 Replies Last Post 7 dic 2012, 14:41 GMT-5
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
3D indeed tend to get very Biiiig and RAM consuming ;)
a few tricks: cqan you use symmetry, to divide our model in 1/ 1/4 or even 1/8 or more ?, can you use 2D as the variations along one axis is small, or do you have a shape very looong, so that the section can be reduced to a point, or do you have cylindrical symmetry, then 2D-axi or 1D axi even is much quicker to solve
Once you have solved for 2D-axi or 2D you can map it back to 3D to give 3D views
Note: COMSOL solves always in 3D, but when you select 2D, the formulas are simplified, and the variables are considered per 1 meter in depth (Z) with no gradients along Z, (or 1 meter square in the section Y-Z perpendicular to the line direction X)
Welcome to the real world :)
--
Good luck
Ivar
3D indeed tend to get very Biiiig and RAM consuming ;)
a few tricks: cqan you use symmetry, to divide our model in 1/ 1/4 or even 1/8 or more ?, can you use 2D as the variations along one axis is small, or do you have a shape very looong, so that the section can be reduced to a point, or do you have cylindrical symmetry, then 2D-axi or 1D axi even is much quicker to solve
Once you have solved for 2D-axi or 2D you can map it back to 3D to give 3D views
Note: COMSOL solves always in 3D, but when you select 2D, the formulas are simplified, and the variables are considered per 1 meter in depth (Z) with no gradients along Z, (or 1 meter square in the section Y-Z perpendicular to the line direction X)
Welcome to the real world :)
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Thank you for your helpful suggestions, Ivar!
I am also wondering when COMSOL will support GPU during the solution process.
I will at least need hundreds of GB's if I make my 3D model close to reality (subsurface
flow and heat transport with boreholes). Sounds like GPU might be able to get me there,
unless I spend >$15K to buy a super computer.
G
I am also wondering when COMSOL will support GPU during the solution process.
I will at least need hundreds of GB's if I make my 3D model close to reality (subsurface
flow and heat transport with boreholes). Sounds like GPU might be able to get me there,
unless I spend >$15K to buy a super computer.
G
Hi
3D indeed tend to get very Biiiig and RAM consuming ;)
a few tricks: cqan you use symmetry, to divide our model in 1/ 1/4 or even 1/8 or more ?, can you use 2D as the variations along one axis is small, or do you have a shape very looong, so that the section can be reduced to a point, or do you have cylindrical symmetry, then 2D-axi or 1D axi even is much quicker to solve
Once you have solved for 2D-axi or 2D you can map it back to 3D to give 3D views
Note: COMSOL solves always in 3D, but when you select 2D, the formulas are simplified, and the variables are considered per 1 meter in depth (Z) with no gradients along Z, (or 1 meter square in the section Y-Z perpendicular to the line direction X)
Welcome to the real world :)
--
Good luck
Ivar