I'm able to get decent speeds from my 7950s (up to 660 kh/s) with CGMiner. That should translate to 750 kh/s for a 7970.
Sadly, identical video cards in different systems require different settings (depending on OS, drivers and possibly hardware). Clocks are deep magic. I get 650 kh/s on one card at 1100 core / 1250 mem and 660 kh/s on another at 1125 core / 1500 mem. The second card won't perform as the first at the same clocks. Increasing the memory clocks of the first card actually slows it down.
These are the settings you have to tune:
- GPU memory speed. Higher should equal faster.
- GPU core speed. Optimal settings depends on memory speed.
- Thread concurrency. Higher is usually better. Should be a multiple of the number of shaders (2048). High TC requires high memory speed.
- Intensity. Higher is usually better.
- Worksize/Vectors/GPU threads. Set to 256/1/1 for 79x0 cards.