Chunking a gpuArray - can it be done?

My question is a little bit theoretical. I'm trying to develop a code to process a signal, therefore using concepts of DSP. Working with gpuArray is great, but I still need to optimize my code to run even faster. I know that MATLAB does the multi-threading intrinsically, but my question is: if I split my signal (it has 0.5-1GB) into smaller arrays (let's say, 1024 samples each, making it a block, so to speak), would it be faster? Or would it be the same if I tried to parallelized this way?

Thanks in advance!

[Edric Ellis]

Could you post a simple example of the change you're considering making?