Regarding the size of matrix that Matlab can handle

How big matrix that matlab can handle? In my project, the program may need to generate and operate on a matrix with size of 50000*5000, and each entry is a double value.

[Walter Roberson]

Would this be a sparse matrix or not sparse?

[liangjian]

Yes, it is sparse matrix. For instance, I need to use a third-party package, which is assumed to handle normal matrix only. Is there a way that I can save this matrix to a sparse matrix, and let it automatically be expanded when used by that third-party package.

Answer 1

[Titus Edelhofer]

Hi,

it depends on version and operating system: on a 32Bit OS this will definetely fail, because the matrix would need about 1.8GB of memory. On a 64Bit machine (with reasonable new MATLAB release).

doc computer

shows the maximum number of elements allowed in an array.

Titus

[liangjian]

It is maybe a genenic question, how to get the approximate memory usage, like 1.8GB, based on the matrix size and element type?

Answer 2

[Andreas Goser]

On 32 bit Windows machines - still the most common platform - a practical size of matrices ends with about 500MB. Always remember that you need memory for the operation and simply loading is not enough.

Answer 3

[Gregory Vernon]

You could look for ways to reduce your grid size. Are there symmetries you can take advantage of? Can you use a skewed or adaptive mesh? Do you NEED double precision?