2009-01-30, 18:27
I'm using an AMD 5000+ at stock, with an NVIDIA 8200 integrated GPU - pretty accessible equipment.
I've had an idea on getting rid of the need to pull the data off the GPU, if that works out it'll be pretty incredible to be honest - i've already got it decoding and presenting to an offscreen surface without the need for the memory transfer, and using that method (which admittedly doesnt perform the final step of 'displaying' the image in XBMC) the CPU usage decoding killa (obviously with no frame drops) is around 10% per core - pretty much what XBMC idles at.
The holy grail is getting the texture to GL without having to pull it onto the CPU, and i'll get there... eventually...
I've had an idea on getting rid of the need to pull the data off the GPU, if that works out it'll be pretty incredible to be honest - i've already got it decoding and presenting to an offscreen surface without the need for the memory transfer, and using that method (which admittedly doesnt perform the final step of 'displaying' the image in XBMC) the CPU usage decoding killa (obviously with no frame drops) is around 10% per core - pretty much what XBMC idles at.
The holy grail is getting the texture to GL without having to pull it onto the CPU, and i'll get there... eventually...