I don't think there is a "best" way to do it.

What i do is:
1. Get the best configuration possible using cuda miner
2. Set memory clock to the lowest possible point (Mine is -502).
3. Increase the GPU clock by 10.
4. Wait 15 mins to see if you have any errors
5. If no errors or driver crashes go to step 3 else go to step 6.
6. If there are driver crashes or errors reduce gpu clock by 10
7. Profit!