I'm just going to leave this here:
Numbers were obtained with one of these, ASIC quality 79.8% (which doesn't appear to be anything special). I used Furmark loading the GPU to 100% in a 1280x720 window, and each clockspeed data point was obtained from the GPU-Z 0.8.2 logfile using the average of at least one minute of sampling at 0.1s refresh rate (so 600+ samples per data point).
I also checked that setting the clock limit to just below that average speed in the results table gave me a stable core clock that didn't boost or throttle during Furmark. The two things that stand out for me after graphing the results are:
- There's a definite curve to the performance drop-off, and at 1000MHz, there's an awful lot (22%) of power envelope that's unnecessary.
- The largest perpendicular distance between the two lines when all axes are plotted from zero is in the 850-900MHz region, indicating that this is (my) Hawaii's most efficient clockspeed, and probably the clockspeed that the 290X should have launched at in a world without the 780 and Titan.
Don't feel you have to plot a graph or sample every single powertune setting, but I guess it would be interesting to see at least two things:
- How little power do you need to run at, say, 900MHz?
- What can your card can do with a 200W TDP instead of the default 290W TDP (-31% power limit)?