Interesting what happened there when you switched the cards. Not sure what that's about. The most important observation I think you could make is ... when you use each card separately, in both slots, do you get the same gpu usage profile in all four situations? If not, then you should start considering either a bad slot, or a bad card, depending on 'where the difference occurs' ... I'm sure you're a good enough troubleshooter to understand what I mean
Those most recent usage profiles definitely look a TON better than the earlier Heaven one you showed us. Not 100% 'optimal' ... these tests really shouldn't show the gpu usage varying QUITE so much, but occasional brief dips during scene changes (esp. in the Valley one) are normal. I'd also like to see the usage running at 97-99% nearly all the time, but it looks like you're running around 95% average, which is very good, just not 'perfectly' in line with I think these cards are capable of doing. It may just be an SLI thing, though, my direct experience with Kepler is limited to my single 670.
EDIT: actually you know what ... your usage level there is actually normal/fine ... your bottom card is running up at 98-99%, it's only your top card that's running at 95%. That's because the max boost on the top card is a higher clock. 1163 vs 1124 ... that's why that's happening. Perfectly normal behavior there.
You might try checking out this thread
as there's tons of graph screenies in there of people's Heaven runs. There's also a Valley benchmark thread
that has lots of graph screenies as well.
I'd probably focus first on really nailing down whether either single card or slot behaves differently from the other with various tests at this point, myself.Edited by brettjv - 3/25/13 at 9:07am