MATLAB Answers

Is the Global Optimization Toolbox GPU compliant?

20 views (last 30 days)
Stefan
Stefan on 25 Jun 2012
I am wondering if my code will benefit from a speedup from a GPU (i.e. wondering if I should buy one), and realized that I don't know if the Global Optimization Toolbox is GPU compliant?
Thanks! Stefan

  0 Comments

Sign in to comment.

Accepted Answer

John Melonakos
John Melonakos on 26 Jun 2012
It is not. Your best bet is Jacket.

  1 Comment

Stefan
Stefan on 26 Jun 2012
Thanks! I suspected as much. Already emailed the accelereyes people :)

Sign in to comment.

More Answers (1)

Walter Roberson
Walter Roberson on 25 Jun 2012
That document shows how the various commands can be used with parallelism options.
I do not know if the GPU can be accessed from multiple workers at the same time (I haven't studied the GPU.) I have, however, read that only one GPU can be accessed per MATLAB instance (parallel workers are separate instances, I gather.)

  3 Comments

Stefan
Stefan on 25 Jun 2012
I am aware of the parallelism options and am making use of them.
From a cost point of view it's much cheaper to get a hold of a GPU that is CUDA compliant rather than trying to build a processor farm, so why I am interested in GPU compliance for that particular toolbox.
Thanks for taking the time to reply!
Stefan
Walter Roberson
Walter Roberson on 26 Jun 2012
Some of the routines (such as MultiStart) wrap around existing commands, with those commands invoking user functions to do the calculations. The user functions can invoke GPU calculations (provided that you have the Parallel Computing Toolbox).
If, however, you are using the options that invoke parallel instances, then I have no information at this time as to whether the parallel instances can simultaneously use the same GPU.
Your question is not clear as to what model of GPU usage you are hoping for ?
Stefan
Stefan on 26 Jun 2012
I am hoping to be able to run patternsearch on the GPU instead of on a few different worker threads. I suppose that's asking a bit much of GPU computing (as it does not support a lot of operations).
Thanks for the reply once again!

Sign in to comment.