Paul, 1600x1200x32bpp with triple buffering would exhaust all local memory (or at least 30MB of it).
But 4x vs. 2x vs. 1x is also about optimized software. At 4x 16 bytes can be transferred in a single clock cycle. If only 8 bytes of data are transferred, 8 bytes of garbage are included. The bandwidth is there, but it is wasted.
For 4x to be measureably twice as fast as 2x, all AGP transfers would have to contain 16 bytes of meaningful data.
But 4x vs. 2x vs. 1x is also about optimized software. At 4x 16 bytes can be transferred in a single clock cycle. If only 8 bytes of data are transferred, 8 bytes of garbage are included. The bandwidth is there, but it is wasted.
For 4x to be measureably twice as fast as 2x, all AGP transfers would have to contain 16 bytes of meaningful data.
Comment