current nVidia and chipset implimentation does not allow for it's advantages to show.
this is ONE reason why DDR SDRAM shows a LARGE improvement in game performance:
data travelling to the AGP bus from the CPU normally has to be seen by the memory as well (don't ask me why, as the concept is totally opposite to CSMA/CD ethernet networks, which have been standard since.. oh 1985?) in current implimentation, however if it was allowed to BYPASS the system memory, it FREES up the memory to do other things, that are important to the CPU. it also would keep the system memory from being the bottleneck in transfers between the CPU and AGP port.
for some calculations:
current SDRAM runs at a normal max at ~1 GB a second correct?
AGP 4X is 66mhz, 64 bit (I think), WITH 4x64bit transfer PER CLOCK. therefor the theoretical max for the AGP 4x port to transfer is 2112 megs/second.
as u can see, if the system RAM must participate in this conversation between the CPU and the AGP port, there is the potential for the system RAM to slow it down, NOT TO MENTION that the system RAM has to be used for NUMEROUS operations for the creation of the T&L data that will be sent BACK, AS WELL AS AI, Physics etc.
IF fast writes was enabled PROPERLY, we would see perhaps half the improvement that DDR SDRAM currently gives us (not a definate number, becuase I don't know all the details).
Facebook
Twitter