Hi everybody.
I have a problem about volume meshing. Today I tried to re-mesh old geometries in order to compare different meshing software, but I can't mesh them. I set the meshing options but something goes wrong and I obtain just a surface mesh. Please, can you take a look? I need an e-mail address to send a simple STEP model as example, here it's not possible to attach anything.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I think it depends on my hardware resources or some troubles between O.S. and hardware (graphic card?). If meshing granularity and max-mesh size are not too high, I can obtain a volume mesh. Above around 1500000, Netgen closes or stops at surface mesh.
Other meshing softwares - like Gridgen - have not this cells limit. I use Ubuntu 10.04 LTS on a little workstation: a Intel Xeon dual-core (2 Ghz), 4 Gb RAM, Nvidia Quadro FX 550.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Oh, 1,500,000 elements? That's quite a lot! I don't have the patience to wait for such a fine mesh, but I wouldn't be surprised if it ran out of memory, especially if its 32 bit.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I think it's not such a big number for CFD simulations… and I guess it's not a general hardware resources problem: I have not the same troubles with other softwares. Why Netgen should be sensible about the 32bit architecture and other softwares not? Did anyone experiment such problems?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Normally, a Linux System running on 32-bit Hardware limits the memory available to a given process to a maximum of 2GB (In some forums I see a limit of 3GB, but I think that is with a special PAE Kernel).
I have experimented running Netgen on 32-bit and 64-bit hardware specifically for the maximum number of cells in each case, and found the following:
1. On a 32-bit System, with KDE running (along with its gadgets, and bells and whistles), and system memory of 8GB, I can create a mesh of up to around 2 Million cells.
2. On a 64-bit System with KDE running (same as above), and system memory of 8GB, I have meshed up to around 8 Million cells using Netgen (but I did not try to go up to the limit….. I think with 8GB RAM I may be able to go up to around 12 Million cells).
I dont know which other memory intensive software you have tried…. maybe the software you tried works with multiple process threads….. in which case you can technically get up to 2GB per process.
Have a nice day!
Philippose
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
1. A 32-bit system has a maximum addressable memory limit of 4GB (global system-wide… including graphics memory, PCI, etc…etc..etc….)… (This comes from 2^32)
2. If you have a graphics card with a lot of memory, the maximum amount of user addressable RAM becomes smaller.
3. For a 32-bit Linux System with a high-end graphics card, you cannot get more than 2GB of RAM per process
4. For a 32-bit Windows System with a high-end graphics card, the general limit is 2GB, but usually you get only around 1.6GB
5. I was checking up Gridgen support forums, and the 32-bit version of the software also has the same limitation, and this limitation does not exist for the 64-bit variant
6. Pure Tetrahedral Meshes require roughly 6 times more memory than pure hexahedral structured meshes. So maybe you were working with either Hexahedral or hybrid meshes in Gridgen because of which the problem never manifested itself in your case.
Philippose
(P.S….. a 64-bit hardware has practically no limit on the addressable memory…. it is 2^64 which is a couple of Exabytes)
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thanks fo your exhaustive reply. Anyway, I use Ubuntu 10.04 with a PAE kernel on my 32bit system, so I really have 4GB. Besides, I think a Nvidia Quado FX 550 is an entry level graphic card, not a high-end one.
After your notes, I think my troubles depend on the 32bit architecture. Maybe other commercial codes (like Gridgen/Pointwise) are optimized and go over open source code limits (like Netgen). Well, I know they are faster and manage a larger amount of cells becouse human and money resources are different.
I must get a 64bit workstation. Thanks for your explanations.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I installed 2 64bit o.s. (CentOS 5.5 and Ubuntu 10.04 LTS ) and I saw the difference at once. I can manage 5M elements without troubles and graphic is also more fluid. The same geometry crashed with around 1.5M cells before.
About that, I have another question. I have no problems with a coarse or very coarse mesh option, but I can't have even a 'moderate' mesh (meshing process stops with surface mesh): what does it depend on? Other user with the same problem?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi everybody.
I have a problem about volume meshing. Today I tried to re-mesh old geometries in order to compare different meshing software, but I can't mesh them. I set the meshing options but something goes wrong and I obtain just a surface mesh. Please, can you take a look? I need an e-mail address to send a simple STEP model as example, here it's not possible to attach anything.
Did it give an error? If not maybe you set it to stop before finishing the volume mesh.
How did you save the output? A volume mesh appears to be a surface mesh in the viewer without turning on the right view options.
See any abnormalities in the surface mesh that might break it, like overly fine regions?
Do any other STEP files work?
You can put a file on dropbox.com and link to it from here.
I think it depends on my hardware resources or some troubles between O.S. and hardware (graphic card?). If meshing granularity and max-mesh size are not too high, I can obtain a volume mesh. Above around 1500000, Netgen closes or stops at surface mesh.
Other meshing softwares - like Gridgen - have not this cells limit. I use Ubuntu 10.04 LTS on a little workstation: a Intel Xeon dual-core (2 Ghz), 4 Gb RAM, Nvidia Quadro FX 550.
Oh, 1,500,000 elements? That's quite a lot! I don't have the patience to wait for such a fine mesh, but I wouldn't be surprised if it ran out of memory, especially if its 32 bit.
I think it's not such a big number for CFD simulations… and I guess it's not a general hardware resources problem: I have not the same troubles with other softwares. Why Netgen should be sensible about the 32bit architecture and other softwares not? Did anyone experiment such problems?
Hello,
A Good Day to you!
Normally, a Linux System running on 32-bit Hardware limits the memory available to a given process to a maximum of 2GB (In some forums I see a limit of 3GB, but I think that is with a special PAE Kernel).
I have experimented running Netgen on 32-bit and 64-bit hardware specifically for the maximum number of cells in each case, and found the following:
1. On a 32-bit System, with KDE running (along with its gadgets, and bells and whistles), and system memory of 8GB, I can create a mesh of up to around 2 Million cells.
2. On a 64-bit System with KDE running (same as above), and system memory of 8GB, I have meshed up to around 8 Million cells using Netgen (but I did not try to go up to the limit….. I think with 8GB RAM I may be able to go up to around 12 Million cells).
I dont know which other memory intensive software you have tried…. maybe the software you tried works with multiple process threads….. in which case you can technically get up to 2GB per process.
Have a nice day!
Philippose
For further clarification…..
1. A 32-bit system has a maximum addressable memory limit of 4GB (global system-wide… including graphics memory, PCI, etc…etc..etc….)… (This comes from 2^32)
2. If you have a graphics card with a lot of memory, the maximum amount of user addressable RAM becomes smaller.
3. For a 32-bit Linux System with a high-end graphics card, you cannot get more than 2GB of RAM per process
4. For a 32-bit Windows System with a high-end graphics card, the general limit is 2GB, but usually you get only around 1.6GB
5. I was checking up Gridgen support forums, and the 32-bit version of the software also has the same limitation, and this limitation does not exist for the 64-bit variant
6. Pure Tetrahedral Meshes require roughly 6 times more memory than pure hexahedral structured meshes. So maybe you were working with either Hexahedral or hybrid meshes in Gridgen because of which the problem never manifested itself in your case.
Philippose
(P.S….. a 64-bit hardware has practically no limit on the addressable memory…. it is 2^64 which is a couple of Exabytes)
Hi, Philippose.
Thanks fo your exhaustive reply. Anyway, I use Ubuntu 10.04 with a PAE kernel on my 32bit system, so I really have 4GB. Besides, I think a Nvidia Quado FX 550 is an entry level graphic card, not a high-end one.
After your notes, I think my troubles depend on the 32bit architecture. Maybe other commercial codes (like Gridgen/Pointwise) are optimized and go over open source code limits (like Netgen). Well, I know they are faster and manage a larger amount of cells becouse human and money resources are different.
I must get a 64bit workstation. Thanks for your explanations.
I installed 2 64bit o.s. (CentOS 5.5 and Ubuntu 10.04 LTS ) and I saw the difference at once. I can manage 5M elements without troubles and graphic is also more fluid. The same geometry crashed with around 1.5M cells before.
About that, I have another question. I have no problems with a coarse or very coarse mesh option, but I can't have even a 'moderate' mesh (meshing process stops with surface mesh): what does it depend on? Other user with the same problem?