How to solve Matlab memory issue

Question

Kavitha Srinivasan 2012-11-21

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/54316-how-to-solve-matlab-memory-issue

Hi all

I am working with huge data file of matrix size 1024*736 and has 367 such files. The approximate RAM required for processing this is 16 GB. I am using Matlab 64-bit version in Linux of 16 GB capacity. But matlab freezes even at 4 GB. What is causing the problem? Is Matlab is not configured to utilise the system memory at all. Are the matlab and system are not working in hands to solve memory problems. How to get rid of this?

Thanks for all your support.

2 个评论
显示无隐藏无

Walter Roberson 2012-11-21

Are you sure about the processing memory? If the matrices are double precision, then the total memory of what you outline is a bit over 2 Gb to load them. What kind of operation do you do on them after they are loaded?

Kavitha Srinivasan 2012-11-21

Yes. the matrices are stored as double and I used FDk reconstruction operation, (that involves weighting, filtering, back projection) of those input files.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Jan 2012-11-21

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/54316-how-to-solve-matlab-memory-issue#answer_66031

在 MATLAB Online 中打开

What is the output of:

[a,b,c] = computer

The memopry can be exhausted by other problems also, e.g. when a pre-allocation is forgotten. Then the finals amount of memory might be some GB only, but temporarily thousands of GB could be allocated. So please post a minimal example which reproduces the problem. Usually rand is sufficient to simulate reading large files.

3 个评论
显示 1更早的评论隐藏 1更早的评论

Walter Roberson 2012-11-22

(FDK appears to mean Feldkamp-Davis-Kress)

Kavitha Srinivasan 2012-11-22

yes. its cone beam reconstruction algorithm

请先登录，再进行评论。

Answer 2

Matt J 2012-11-22

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/54316-how-to-solve-matlab-memory-issue#answer_66147

编辑：Matt J 2012-11-22

在 MATLAB Online 中打开

To do FDK reconstruction, you don't have to load the entire data set into memory at the same time. Traditionally, it is done by processing one projection view at a time. If your 367 files each contain a projection view, why not just load one of them into RAM at a time, do the processing needed to update your image volume, ad then throw it away?

Also, how do you rationalize

memo = 8*(4*nx*ny*nz + 2*nu*nv);

If your image volume is being held as 4-byte singles, shouldn't it just be

memo = 4*nx*ny*nz;