This is faster due to the fact the modern graphic(dedicated) cards have hundreds of GPUs executing in parallel .The only issue is the bus between CPU and GPU. the transition of the data to the card and from it is what usually prevents to get the best out of this approach.But still I believe if you architecture the CPU (client side) code the right way you will gain much from it.I did not try it.I am learning PB3D now because I am not going to suffer doing such a stuff with opcodes.As I wrote today earlier "number crunching "(term to what you want to do ) was already done in PB2d .There is an article on Adobe PixelBender page describing how to send computations to PixelBender and back.Read it,I think it will be helpful .:)
On Sun, Mar 20, 2011 at 12:28 AM, Arkadianen <[email protected]> wrote: > This matter is very importnat, because collision detection takes lots > of CPU. Is GPU able to calculate it faster than CPU? Maybe we could > even write shaders for path finding or other games tasks like AI ? > Did anyone try it? > Having links? > > On Mar 19, 11:11 am, Michael Iv <[email protected]> wrote: > > I am still learning the API but here are some things I can tell you : > > Yes you can send the data to the gpu to be calculated . For this you > should write a shader program which would get input of soma data and return > the calculated result . I think the best way to do that is via pb3d so you > don't have to mess around with opcodes. Also this thing was already done > with the regular pixel bender. There is an article in pb adobe web page > about number crunching using pb shaders. Just take a look at it .btw , the > industry standard physics engines like PhysX and Havok run on GPU :) > > > > Sent from my iPhone > > > > On Mar 19, 2011, at 1:34 AM, Darcey Lloyd <[email protected]> > wrote: > > > > > > > > > I have just been wandering through some gpu collision detection > articles and was wondering: > > > > > Q1). Is there any way to access the GPU for collision detection > queries, freeing the CPU from testing? Or when we do a comparison between > two vertices does molehill automatically get the GPU to do this? Is this > possible via stage3D / Context3D? > > > > > Q2). How much access do we have to the GPU? > > > > > Q3). Would it be possible to have more access to GPU functionality via > pbj objects (Pixelbender)? > > > > > D > -- Michael Ivanov ,Programmer Neurotech Solutions Ltd. Flex|Air |3D|Unity| www.neurotechresearch.com http://blog.alladvanced.net Tel:054-4962254 [email protected] [email protected]
