2006b runs M-file faster than 2011a

Hi, A M-file runs in Version 2006b in 22seconds whereas the same file takes 4 times to run in 2011a,more than a minute.The file will be 4000 lines approximately.Why?

[Walter Roberson]

Which kind of processor do you have? In particular does it happen to be an i5 or i7 ?

[Andreas Goser]

Walter, do you have a specifc issue in mind? I actually have right now a customer with a i5 to i7 system migration issue.

[Jan]

@Andreas: There have been some reports of slow i5/i7 interpretation on 64 bit systems in this forum.

Answer 1

[Jan]

Without seeing the code it is impossible to guess the reason. Speed issues on modern processors and 32/64 bit versions have been reported repeatedly in this forum, see: Answers: Search slow+i5, e.g.:

• Answers: computing times differences.
• Answers: slow-script-execution-with-new-laptop

Answer 2

[Daniel Shub]

There is a bench type function on the FEX: href=""<http://www.mathworks.com/matlabcentral/fileexchange/11984-benchmark</a>> that can be used across versions.

I thought that TMW made the bench test from different versions available, although I cannot find them.

Answer 3

[Andreas Goser]

This answer is about the first steps to indentify reasons. First of all, it is assumed both installations are on the same machine. Otherwise the comparison is not so valid.

1. Run

bench(5)

on both installations and see if the results differ a lot and if yes in which category most. You may find information about a certain slowdown in the community or through Technical Support.

2. Run the profiler File -> Desktop -> Profiler or through the PROFILE command for both installations and see whether there is a certain line of code that dominates the slowdown. Again yu may find information about a certain slowdown in the community or through Technical Support.

[Friedrich]

Beware of bench, because the size of the tested problems can change between releases. So it will most likely show that 6b is a way faster than 11a when using bench.

[Jan]

@Friedrich: Thanks for mentioning this "surprising" behavior of BENCH. It has been a bad idea to modify the problem size of a function, which should help to compare different platforms.