Help Me Overclock my Radeon!?

[pap]

128bit memory interface is not faster.. it just allows more bandwidth availability... which allows more data to be worked on @ the same time...

the reason the SE is slower is the memory/core is slower... much slower than the XT's AND the memory bandwidth is more limited...

if you can work on more data at one time than the end result WILL be a faster output

 

[Sazar]

if you can work on more data at one time than the end result WILL be a faster output

possibly... yes

but when you have more data being worked on @ the same time by a 64 bit and a 128bit memory controller... if there is a need for more than 64bit memory interface (depends on the info remember) you are dealing with a scenario where the 128bit will likely be the same or faster.. however if the requirement is for 64bits of data accuracy you are likely to see a drop in performance vis a vis 64bit memory..

the two most likely ways to boost performance are clock speed increase or work done per clock cycle..

you are right about the memory interface being a factor however... I just don't think its a big factor when it comes to performance @ the levels expected for the cards that are 64bit v/s 128 bit in the same model category...

 

[champ2005]

Well i remember when the ill-fated Geforce FX5800 Ultra came out, it had a much faster memory speed than the 9800 Pro, but it only had a 128-bit memory interface vs the 256-bit memory interface of the 9800 Pro. This gave the 9800 Pro quite an advantage when the graphics/scenery became much more complicated (which consumes more memory) (im talking DX8.1 here).

Its a bit like 56K internet vs broadband

Both will download a 5K file in less than 1 second. (really bad graphics like cs )
But broadband is much faster for a 50MB file. (more complex scenery like code creatures)

 

[Sazar]

the comparison you are talking about is between the 5800ultra and the 9700pro...

the difference was not only the memory but also the pipelines... 4x2 v/s 8x1 and that makes a difference @ high LOD and IQ... not only that the architecture of the r3xx allows for more work to be done per clock cycle...

keep in mind that in non stressing tests (not requiring extra pixel-pushing power) the 5800ultra was generally able to match or exceed the performance of the 9700pro...