2D FFT performance on GTX Titan

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi guys,
i know FFT is a useful performance benchmark, and GPU do it very well expecially with more then 1024 point.
I would like to know processing time for a 2D FFT with 4096x4096 point because i need to use it in a real time application.

Does anyone have useful information about this?
Thanks a lot.

2

I’m using a 3D FFT and I assure you it’s really fast. With almost 2 M elements it takes ~0.7 ms to perform each of these functions:

cufftExecR2C()
cufftExecC2R()

So, knowing that FFT is O(n log n), you can estimate how much it will take you. Although, the best way to know it is to run your own benchmark.

3

Thanks for your reply, of course i will know the exact latency after having tested the benchmark, but at the moment we are still waiting for the Titan from the dealer…

4

2M = 2 million of elements. In my case, 2 million of floats.
To estimate the processing time in your case, having O(n log n) complexity, you can use my time: 7·10⁻⁴ s

You have a conversion factor: C = 7·10⁻⁴ / (2·10⁶·log(2·10⁶))
so, if I’m not wrong,
~16·10⁶ · C = ~6.4·10⁻³s = ~6.4 ms

As I told you, the best way to know it, is to make your own benchmark and see the results because there could be another factors affecting your performance.