feinicks Wrote:You are missing the point. I am not saying that at this moment, GPU as they are, can replace the processing prowess of CPU. Processing is inevitably directly related to the no. of transistors available. GPUs now exceed the complexity of modern CPUs in terms of absolute transistor count. And like CPUs, they're becoming programmable-- it's possible to harness all that graphics power to do something other than graphics. So while they may not be fully optimised for serial processing as yet, they already have started to move in that direction. Further, customised hardware in a GPU allows Stream processing. If anyone doesn't know, stream processors are extremely powerful floating-point processors able to process whole blocks of data at once, whereas CPUs carry out only a handful of numerical operations at a time. Wee've seen CPUs implement some stream processing with instruction sets like SSE and 3DNow!, but these efforts pale in comparison to what custom hardware has been able to do.

CUDA is one step in that direction. My point being that in the future, this will be something that will be very interesting to observe.

I seriously doubt GPUs will ever beat CPUs in terms of serial processing, with the current way things are going.  Of course, with Intel and AMD planning to integrate graphics processing onto CPUs, things may change direction, but on the current path, a CPU will always run a serial task much faster than a GPU can.  Don't confuse raw processing power with ability to harness it - from a programmer's perspective, mass parallelism is very difficult to achieve with many applications, so this isn't just really a "wee don't have the applications today" issue.
I'm not digging CUDA much.  For one, ATI doesn't support it, and unless nVidia squeeze them out of the market, it's going to be difficult to adopt to such a solution.