Hi guys;

I am using Gambit on a supercomputer (unix) and running out of memory while meshing a 6 ft circular duct (6 inch. dia) with 0.1 inch. 'stairstep'.

I submitted a batch job specifying 4 CPUs (parallel) and 8gb of memory, however the process was killed after 7000 something seconds. Out of memory!!! (how come???).

The system technician told me that I probably underspecified number of CPUs and memory in Gambit (What???).

Anyone knows how to specify memory usage in Gambit (if there is any)?

Thanks in advance.

Ahmet

By the way, I specified the 4 cpus and 8gb memory outside gambit. The batch file I submitted is as follows.

-----------------------------

#PBS -N duct

#PBS -S /bin/tcsh

#PBS -j oe

#PBS -l cput=10:00:00

#PBS -l ncpus=4,mem=8gb

cd /scratch/a0u8545/060102

module load gambit

gambit -inp 6inch.jou

qstat -f $PBS_JOBID

-----------------------------

Gambit doesn't run in parallel, no matter how many processors you define in your LSF (I'm assuming it's an LSF or some variant based on the input scripts... I use similar stuff here at work). And depending on how your LSF is set up, a single process is often limited to 2Gb of RAM. Some LSFs let you unset the per-process memory limit (The per-process memory limit is different from the memory allotted per processor... like when you're running 3 or 4 programs on your desktop at home... EACH of those programs is limited to a maximum RAM usage of 2Gb by the operating system no matter how much RAM is in the computer).

A quick estimate of your mesh size though... 6" diameter x 6' long pipe... that's a volume of ~2035in^3... a .1" hex mesh would result in about ~2million elements. You shouldn't be hitting a memory limits with that small a mesh. I've meshed 8 million tet elements with complicated geometry without any memory limits. I wonder if there's something going on in your Gambit journal that's causing the problem.

Good luck, Jason

Hi Jason,

First of all thanks for the information on the processor and the RAM. That means I need to deal with less variables.

The journal file should work ok because I tried it with simpler geometries.

I learned that there is a program variable that you can set the memory

"global..gambitmem"

However, I couldn't get it to work so far. Maybe I am typing it in wrong. It also said gambitmem is not defined.

The following link has the information on that if you are interested.

http://zsc.zcu.cz/sw/manuals/gambit/...htm#cr05010204

Thanks again.

Ahmet

[akshayygarg]

HiI am running out of memory 2gb although i have ram of 4gb and space more than 40gb on disk. plz need help. program icrashin don again and again.Akshayy Garg

[-mAx-]

you are trying to mesh something which ask lot's of memory.
Decrease the mesh size, or split your domain

[akshayygarg]

Hi

Thanks for the earlier reply. Actual problem is this.

I have two volumes:
Vol1. Cylindrical - 340mm dia and 600mm height
Vol2. Cubical - 200X200X0.3mm

100 volumes of Vol2 are kept in middle of Vol1. Now these 100 volumes are subtracted from Vol1. The new volume so formed has to be meshed using GAMBIT. But due to very fine volumes i am not able to do so.

Kindly Help.
Akshayy Garg
(IIT Delhi)

[-mAx-]

split your domain (by dividing it, you will mesh subvolumes which will ask less memory)

[Daniel Tanner]

Gambit is 32-bit and not parallelised. This means that a gambit process can access a theoretical maximum of 2^32 bits of RAM ~ 4Gb. However, these are limited by the platform you are using (Windows/Linux....) to ~2Gb. If you have memory issues: reduce mesh size, split mesh into smaller volumes or use TGrid (which is 64-bit).

[-mAx-]

but gambit for linux amd64, should be 64bit.
I assume it can handle more than 2GB, right?

[Daniel Tanner]

Sorry, Max is right. I hadn't realised that the 64-bit version was available. However, the rest still holds. 32-bit => ~2 Gb but if you have access to a 64-bit computer you will not have any problems. If you are stuck with 32-bit then split domain, reduce mesh size.

[Geon-Hong]

Jason/

Thanks. Your reply helps me to solve my problem.

Every time I meshing a 3D wing I encountered memory problem of GAMBIT, which was terminated due to excess of memory of 2GB.

It was totally journal file problem!

I removed the journal file and finally completed meshing the wing.

[Ghayoumi]

i have this problem too, but i can't solve it, please help me .....
please enlarge it ......

[emreg]

i have these system specifications:
- i7 2600 intel 64x
- 8 gb ram
- 4 processors

I am trying to mesh a volume which has nearly 4-5 million cell. However suddenly i get a warning message "out of memory".

But on the another computer (2 gb ram, dual core) i can manage 4.5 million meshes in the same conditions.

Where is the problem?

Thanks

[Ghayoumi]

Hi,
To solve this problem you must use a Linux version of gambit!
gambit is a Linux program originally.

[emreg]

hi, how many cells can be meshed using Linux gambit?
It can be 10 million ?

[ghost82]

Originally Posted by emreg

hi, how many cells can be meshed using Linux gambit?
It can be 10 million ?

I successfully meshed 8 million cells on the fly using gambit linux 64 bit (I used 32 gb of ram).

Daniele

[emreg]

are you expressing that the meshing of 8 million on the one time (not sequently) is possible using linux ?

[Far]

Originally Posted by emreg

are you expressing that the meshing of 8 million on the one time (not sequently) is possible using linux ?

Yes it is possible. Any how Tgrid and ICEM CFD are capable of handling of over 40 million mesh size on desktop computers...

[ghost82]

Originally Posted by emreg

are you expressing that the meshing of 8 million on the one time (not sequently) is possible using linux ?

Yes, that is what I wrote.
This is possible if you have enough ram, a linux 64 bit operating system and gambit 64 bit.

Daniele