Good morning everyone!
How do you explain the out-of-memory status in this case (see photo) when Task Manager shows that only about 9 GB of 32 GB are used?
Thank you to everyone who can answer this.
ACCEPTED SOLUTION
Accepted Solutions
Hi,
Mathcad 14/15 is a 32 bit software package so can address memory of 4GB max only.
It looks like only a 34 x 34 matrix is being dealt with you should not be getting memory errors. You are trying to do the determinate solve through symbolics not numerics? The symbolic determinate for a 34 x 34 matrix written out using the 15 digit floating point values is a huge thing.
Using numerics no memory errors:-
Prime is 64 bit and does not have this limitation.
Even Prime has its limits for instance a dense square matrix of 46329x46329 breaks the bank on my 128GB machine.
Using x = lsolve(A,b) in Prime is limited to about 16600 x 16600 elements for A. Bigger than this you need Cholesky() or QR()
I deal with problems in Prime with larger sizes than this using sparse matrices saved to binary files. Then with C++ sub programs that read the binary files to solve the equations. Prime spawns the C++ programs. C++ programs use the NVIDIA GPU through the new cuDSS. Matlab solves in about 40 seconds but the GPU using cuDSS takes 5 seconds.
Cheers
Terry
Hi,
Mathcad 14/15 is a 32 bit software package so can address memory of 4GB max only.
It looks like only a 34 x 34 matrix is being dealt with you should not be getting memory errors. You are trying to do the determinate solve through symbolics not numerics? The symbolic determinate for a 34 x 34 matrix written out using the 15 digit floating point values is a huge thing.
Using numerics no memory errors:-
Prime is 64 bit and does not have this limitation.
Even Prime has its limits for instance a dense square matrix of 46329x46329 breaks the bank on my 128GB machine.
Using x = lsolve(A,b) in Prime is limited to about 16600 x 16600 elements for A. Bigger than this you need Cholesky() or QR()
I deal with problems in Prime with larger sizes than this using sparse matrices saved to binary files. Then with C++ sub programs that read the binary files to solve the equations. Prime spawns the C++ programs. C++ programs use the NVIDIA GPU through the new cuDSS. Matlab solves in about 40 seconds but the GPU using cuDSS takes 5 seconds.
Cheers
Terry
