Reaching the theoretical 'max' speed in opengl
by
, 29-04-2011 at 07:47 PM (6227 Views)
As we all know, if we could hit the max speeds of our hardware, we would be pretty happy, but we never do (zengl gets pretty darn close though)...
Finishing particle effects let me realize this, big time. Its pretty much the same performance hit I saw when I had tiled maps and others... It got me thinking though: what if you didn't render everything everytime but say you did
My next (current) POC idea is simple: render a commonly used frame and reuse it. Basically, I'm attempting to find patterns in draw jobs, analyze those for similarities dynamically and redraw just what needs to be. So basically, that is one very fancy clearcanvas() procedure and some heavy 'do we need to draw this'.
So far, its marginally faster (about 2-6%) - at the expense of 350mhz cpu power but I think I can optimize that a fair deal. However, everything I optimize reduces CPU and GPU usage, but increases memory usage (darn).. Although for now an extra 2 meg is not going to hurt (although its 10%) as I'm thinking of capping it to whatever the programmer specifies...
Stay tuned for the next weird and wonderful idea
PS: Is this the third day I've posted in a row