Best combination for a dual gpu set-up for CUDA heavy calculations


I need some help with setting up a working computer for computational physics. I want to have a dual gtx set-up. What would you recommend? I was thinking to get a 590 gtx, but it seems that they are all sold. the other option was 2x 580 or 2x570. I have no experience with the 6xx series. Some of my programs use heavily the double precision (like lots of double FFT), while others can be divided ( any calculations in single precision and only the final reduction in double, mainly MC and MD simulations). I would like to hear your opinions on this.

I also have some concerns. Some people reported repeated memory errors, many of which can be associated with the overclocking of the memory. How do I know which producer does this?

The standard clock frequencies for NVIDIA GPUs can be found in the specifications listed on NVIDIA’s website. For example, for GTX 680 and Tesla C2075: