So, I spent $400 a couple of days ago on a 7970 GPU thinking that it would run circles around my 570 and 260 because of the increased double floating point performance. All three cards are in one computer. With only one E@h workunit running per card the 7970 doesn't do anything. If I increase it to two workunits per card the 7970 runs around 75% but still takes over an hour to do one workunit. The 570 takes 40 min and uses less of the CPUs. If I increase to three workunits per card the 7970 is running 80% plus but then I run out of CPU cores and the 260 and the 570 chokes. Right now I am only interested in doing einstein@home on the GPUs. I though about taking the 260 out and putting it in another computer that has another 260 but this still doesn't solve the problem. Once I change it on the Einstein@home preferences the 2 260s in the other computer will still choke. Is there a way to wrote a config file for one computer that will assign one job to the 260, two jobs to the 570 and 3 jobs to the 7970? And even if that is possible, is the 7970 the best card for this project. If the Einstein code will not be optimized to take full advantage of the 7970's superior floating point performance should I just return it and get another 570 or 580. They are about $100 cheaper anyway.

Ask the guys at Einstein.

Double precision isn't a matter of 'optimisation', it's a question of whether their mathematical search algorithm requires the extended precision. From what I understand, it doesn't - so it would just be a waste of their programming time to code for a DP card, especially since that requirement would exclude a lot of lesser cards from the project.