Someone experience with this topic yet? We’re developing a web server application in which we accelerate the rendering already with 7900 GTX cards. How ever about 70% of the graphic generation time is currently spent in the png-compression routines using the png/z-lib.
I unfortunately didn’t find the time to start experimenting with CUDA yet and as far as I’ve informed myself till now CUDA promises a very big performance increase for simple functions such as transformations etc…
So my question:
Is it possible to write a CUDA based PNG compression algorithm to reduce our current time of about 5-10ms (dependent on complexity) for a 256x256 big graphic to lets say 1ms?