PC requirements for running multi-Objective optimization algorithms

Hello friends

What type of PC's are best for running a multi-objective optimization algorithm for a complex system? The normal PC/Laptop will take around 4-5 days to complete 100 Iterations. Any special hardware requirements?

[Walter Roberson]

Would the calculation benefit from running part of it on a GPU? If so does it need single precision or double precision?

[Prana]

Thanks Mr Walter

It would be certainly better if it is running on a GPU and alos it needs double precision

Answer 1

[John D'Errico]

Sorry, but this is a question that is far too vague to have a useful answer.

You tell us only that as written by you, your code on your current PC can compute 100 iiterations in roughly 4-5 days. So around 20 iterations per day, so a little over an hour per iteration. But MATLAB does not fail if you might do so, so the problem is not strictly memory limited. The only limitation is time. But we have not been told how well your code is written, so it might be complete crap, and the only reason it uses so much time is poor coding skills. That is entirely possible, since you seem to be not strongly knowledgble in this respect. Or maybe not. We don't know. Is it code written by a student? (Often the case. In that case, poor code is entirely possible. Again, we don't know.)

So what requirements are needed? As fast a PC as you can buy. Or maybe the problem really is memory limited, because the problem is really big and it is sucking up all of the memory you have, and MATLAB is having a hard time there. So maybe just a big PC with lots of memory is the requirement. And that may be again, just poorly written code. We don't know. It might be just the need to learn to use the parallel processing toolbox, (After you buy it, not cheap though, and you would need to learn to use it properly.)

Or maybe the only requirement is to just accept things as they are, and you just invest in a coffee machine, and sit down and read/re-read the complete works of Shakespeare while you wait.

What is necessary?

First, spend some time with the profiling tool in MATLAB. Learn exactly what it is about the code that makes it slow. Are you using a lot of memory? Are matrices and vectors not preallocated properly as is suggested? Are there bottllenecks in the code that would tell you if more speed would help? What are the bottlenecks, and what can be done to avoid them?

Before you can possibly know what hardware is needed to run your code faster, you need to know why it is slow. Unless of course you are just willing to throw away a lot of money, for possibly the wrong reasons. A little time and effort invested upfront can replace a lot of money thrown at a big computer. Or, you can just buy the big, ultra-fast computer, and hope that solves the problem. I would note a basic rule I developed many years ago about computing: problems always expand to be just as large as your computing capacity allows, and then just a bit larger.

[Prana]

Thanks Mr John