to which level should the cards be overclocked? keep fan settings at default for that ? maybe have some kind of temperature limit ?
I know this would probably be a massive pain in the ass at first, but hear me out.
I would try to find the sweet spots for clocks/power consumption.
For instance, if you look at
this chart on Xbit you see Cypress will clock to 800mhz before starting to sip more power, and then jumps off a cliff at >925mhz. (Hey, what are the 6800 series clock speeds again?
) I've noticed this is a similar trait with the other 40nm products, including nvidia's (GTS450's default clock for instance, and what you see high-end AIB partners clocking them at fit these parameters, as does the gtx460 in this review).
For me personally, I don't care if a core will go 5% faster if it uses a ton more power, but at the same time knowing the reasonable upper limits is useful in comparing the absolute value of different products, or understanding which way an AIB bins their products (for power consumption or speed).
I understand that doing this for a new core, or especially a new process technology would be difficult. Perhaps just do it once when a new core configuration is released, or find a reasonable control voltage for best trade-off and best performance/watt across a certain node, and then compare all lines across that control.
It would also be nice to know which if any voltage controller is built into the card (ie Volterra VTxxxx) as some are allowing new features (as you may have seen MSI incorporate into Afterburner recently) Some don't advertise having them as prominently, but that doesn't mean they still don't work!
I know it may sound asinine to ask for any or all of that...but hey, I had to try. I think many people would find it an invaluable resource.