Give a few examples of COMPUTATIONAL apps if you don't mind.
Seems that people are saying the core is unimportant, while the
shader is.
Ah, AMD Stream.
There are few programs that run AMD Stream, but I can use for
NVidia's CUDA.
I have a 9500GS. Card sucks, memory bandwidth sucks more. I used
Badaboom to test out my overclocks (Once deemed stable) and just by
overclocking the memory to its limits (DDR2 @ 1200MHz) I gained
quite a fair bit of speed. Cut transcoding time by about half a
minute, which I deem pretty good for a sucky ass card. Other
computational apps?
Folding @ Home. I have both a CUDA and a Stream version. Using
the CUDA version on said 9500GS, I managed to accomplish more
"tasks" overnight with OC'ed memory than without.
But gaming? Nah. Sucky ass card couldn't keep up.
For NVidia, overclocking core speed by 6% won't really give you
a 6% increase in speed. Maybe 2%? But overclocking shaders will
give you the biggest increase. 6% OC might net you 5% boost?
AMD locks both the core and shader speeds of their Radeon cards.
So you increase both.