Tweet
It's a little unfortunate that these results are carefully (carelessly?) done so as present nouveau in the worst light possible.
The real issue is that on all but the GTX 680 board, memory reclocking completely fails. So you have a high core speed and the original lowest memory clock. This yields totally crap performance (as can be seen).
One would likely get better results by first switching to a middle perf level (usually called "0a" on such GPUs), which should allow the memory clock change without the voltage change, and then try to flip to 0f. The memory reclock will fail again, but at least you'll have a middling memory clock instead of the absolute lowest.
But realistically, any user seriously considering nouveau for gaming would never use upstream. They'd use a tree with Karol's reclocking patches, where every Kepler GPU should behave roughly as the GTX 680 one does - about 60-80% of blob performance, depending on the application.
The real issue is that on all but the GTX 680 board, memory reclocking completely fails. So you have a high core speed and the original lowest memory clock. This yields totally crap performance (as can be seen).
One would likely get better results by first switching to a middle perf level (usually called "0a" on such GPUs), which should allow the memory clock change without the voltage change, and then try to flip to 0f. The memory reclock will fail again, but at least you'll have a middling memory clock instead of the absolute lowest.
But realistically, any user seriously considering nouveau for gaming would never use upstream. They'd use a tree with Karol's reclocking patches, where every Kepler GPU should behave roughly as the GTX 680 one does - about 60-80% of blob performance, depending on the application.
Comment