Hi,
My new Apple laptop has a GeForce 8600M GT graphics card, which according to specs has some new nifty stream processors. I’m interested to do a spot of parallel computation work on them via JOGL. I’m looking for some pointers or examples to get started with. Or even information if this is possible with the current version of JOGL.
Thanks
Peter
OK, well that’s quite a bottle of worms. Perhaps the single best one stop shop is http://gpgpu.org/
But yes GPGPU can be as complex as the effects of batch size and coherence on performance, or as simple as treating textures as 1d, 2d, or 3d arrays.
Is there anything in particular you are interested in doing?
[quote]Or even information if this is possible with the current version of JOGL.
[/quote]
GPGPU with JOGL is the exact same as GPGPU using plain-ole OpenGL, only difference is that it can be easily run on any platform. Now with that taken care of, you need to start looking into how GLSL works. Usually GPGPU work gets done in fragment programs but there is nothing stopping you if either vertex programs or geometry programs location in the pipeline better suites your work.
I thought that fragment shaders were usually used because they generally had faster (and more) fragment processors. Also, I think fragment programs getter slightly higher efficiency when doing texture access.
Yes older hardware had far fewer vertex shader units than fragment shader units, and typically texture fetches were unfiltered and slow. That all changed with the introduction of unified shader GPUs; Xenos, G80, R600, Broadwater. Now vertex processing can utilize all the shader units until the triangle setup rate is saturated. Similarly, texture lookups are now done the exact same way and with the same hardware regardless of whether it’s vertex, geometry, or a fragment program. The only downside I’ve heard of is that the batch size for vertex programs have grown to be or at least close to that of fragment programs, which may/may not have an effect on performance.
Oh and by the way, sorry I haven’t had an opportunity to look at your code yet, though I should have an opportunity to check it out this week.
Thanks, the GPGPU link was useful and I’ve ordered the GPU gems 3 which someone else suggested as a good starting point.
Peter