>>> Wonder if there's a sound card out there that has a programmable DSP
>>> that's got enough parallelism to add as much realism to your simulation
>>> as possible.
>>
>> I imagine a GPU would be quite good at this sort of thing.
>>
>
> I suspect you're right. It has built-in support for parallel operations.
> Can you repurpose a GPU for that sort of thing?
A GPU is basically very fast if you want to perform exactly the same
function on big arrays of data. Nowadays there are very few limitations on
what you can and can't do, and the few remaining are easily got around by
using temporary arrays and multiple functions.
Even if your algorithm apparently doesn't work on huge arrays of data, it
can often be forced to do so in order to get a big speed up. eg With an IFS
program that transforms one point repeatedly to generate a picture, on the
GPU you can transform millions of points in parallel and you get the same
result.
Post a reply to this message
|