Computing times differences between computers

Dear reader,

Together with a friend we developed a certain MatLab program. When we run it on the university computer (2,8 GHz dualcore Intel Celeron with 2GB DDR2 memory) the program is finished in about 1000 seconds. This isn't that bad, we are aware of that.

However, I would have expected my own computer to greatly outperform this time. Phenom II 955 (3.2 GHz Quadcore) with 8 GB DDR3) As you may expect by now, it doesn't. It takes 2000 seconds to finish the exact same program with exactly the same data. My pc only uses about 25% CPU and 1GB memory for MatLab.

How can this be, and how can I solve it?

Greets, Marcel

Additional data:

- At home: Windows7 64bit

- University: XP 32bit

- MatLab at home: 7.13.0.564 (R2011b), 64bit

- MatLab university: 7.10.0.499 (R2010a), 32bit

- Core utilization university pc: 50%, 400MB memory usage

- At university: data is loaded from/saved to a network drive

- At home: all files on C:// drive where MatLab is also installed

'not sure' because i'm home now and can't check will come back on that tomorrow if not resolved

[SCd Merge]

I'm not sure about the university pc, but i'm quite certain it runs on XP 32-bit. My own computer is on 64bit Windows7.

[/merge]

Answer 1

[Sean de Wolski]

Is one 32bit and the other 64? If so, it is expected that the 32bit system will run faster.

Try running bench on both machines to see what specifically takes longer.

[Marcel]

When running Bench here it a relative speed of 56? I'm not sure what that means, but it's among the top best systems it's compared to...

Answer 2

[Image Analyst]

All those cores may not be busy. Do you have the parallel processing toolbox and did you program it with that in mind?

[Marcel]

We did not use that no... We also don't have time to do that anymore so that badly enough is not an option.

[Image Analyst]

Not sure what that means but it sounds like you're complaining about it taking too long but you don't have time to fix it by using better programming methods to take advantage of additional cores in your computer, even if the faster execution time would save you more time than the time to fix it. Hmmmm... Well I guess you're willing to just live with it then, and consider this just a curiosity/puzzle.

Answer 3

[Jason Ross]

Missing details:

• Operating system on each machine? Sounds like XP (32-bit) and Win7 (64-bit)
• 32 or 64 bit MATLAB? (you could be running 32-bit MATLAB on your 64-bit system, which uses the Windows-on-Windows (WOW) layer)
• MATLAB release on each machine?
• Core utilization on the Celeron when you run it?
• Are you dependent on some resource that will be slower on your machine (like a network drive or database that takes longer to give results to your machine because it's going over a more latent network link)

[Walter Roberson]

Right, especially with the reports about R2012a being slow on i5 and i7 for 64 bit versions of MATLAB: we need version information.

Answer 4

[Jan]

• On both computers only one core is used.
• The processor of the computer at home works 10% faster.
• The memory is not the limiting factor, if the program needs 400MB only. Then the modern memory manager of Windows 7 does not matter (much).
• Loading data through the net is usually much slower than accessing a local hard drive. Most likely the file access is not the bottleneck of the program.

Conclusion: Matlab 2010a/32 is twice as fast as 2011b/64 for your problem. This can have several reasons. E.g. if all file accesses ate controlled by an anti-virus program on the home computer. Use profile to find the bottleneck(s) and post the relevant commands to get further advices.