Who fast is the FFT with the new GTX680

At the momenten I use a GTX570 GPU and then CUDA 4.0 compiler. In the fututre I will buy a Geforce GTX 680. Is the floating point (32Bit) FFT performance better with the GTX 680 ?

http://blog.accelereyes.com/blog/2012/04/26/benchmarking-kepler-gtx-680/

There is only a Benchmark for 2D FFT. At the moment I use a 4096x1024 FFT.

Is ist better to buy a GTX 680 with a core frequency of 1200MHz or is GTX 680 with 4GByte RAM better for FFT?

I would go for the 4 GB RAM, but thats mainly because I use 4D FFT’s…

How can I see how much memory are in use if I calculate the FFT?

I will calculate the 4096x1024 FFT (single float)

You can use cuMemGetInfo, before or after the CUFFT call. I don’t know if it’s possible to know the memory usage DURING the CUFFT calculations.

http://nasutechtips.blogspot.se/2011/02/how-to-get-gpu-memory-size-and-usage-in_13.html