Originally Posted by xorbe
As a design engineer I just have a hard time imagining what's going on here. A 16.7M color screen at 1920x1200 is just under 6.6 MB of data. My old 2.8GHz A64 (single core, agp 7800gs) has 7.8GB/s memory bandwidth (dual channel DDR500). AGP 4x/8x is 1-2GB/s.
Recent PCIe is 8GB/s, and DDR2-1066 is 8.3 GB/s per channel.
Where are the cycles/bandwidth being spent on X + 8/9 series cards that it can't compute 6.6 MB of 2D data in less than ~8 ms? It should be able to composite ~4 full 1920x1200 frames in that time, in software.
the problem is how it computes it (basically it isn't) and where the data is going. I suspect the reason vesa and nv are so good at this compared to nv is that acceleration is done solely by the cpu. Nvidia does not accelerate everything so when an operation it cannot accelerate appears it has to pass it off as software incurring a huge peformance penalty.