which FPGA chip to use for FFT?

Hello,

I have made a program which calculate FFT of two images (1024X1024) and
then convolve them. I wish to know which FPGA chips would be best for this
task, I want the result of the algorithm in real time (more than 30hz). It
takes around 10 seconds on my machine(AMD 1.8 Ghz Dual Core) for just one
convolution of both the images. I am thinking to use xilinx XC3S500E (500k
gates). I have no experience in FPGA, if anyone can help me i would really
appreciate. 

Thanks

On Oct 1, 5:53&#4294967295;pm, glen herrmannsfeldt <g...@ugcs.caltech.edu> wrote:
> me_rythm wrote:
> > I have made a program which calculate FFT of two images (1024X1024) and
> > then convolve them. I wish to know which FPGA chips would be best for this
> > task, I want the result of the algorithm in real time (more than 30hz). It
> > takes around 10 seconds on my machine(AMD 1.8 Ghz Dual Core) for just one
> > convolution of both the images. I am thinking to use xilinx XC3S500E (500k
> > gates). I have no experience in FPGA, if anyone can help me i would really
> > appreciate.
>
> You would do better to post in comp.arch.fpga. &#4294967295;(added)
>
> How many bits per sample? &#4294967295;Are you using the SSE feature
> of the AMD machine? &#4294967295;That lets you process multiple 8 bit
> values in one instruction, but might require assembly coding.
>
> I suggest looking at the systolic array architecture as
> it works well for many problems like this. &#4294967295;You don't really
> need FPGA experience, but digital logic experience helps a lot.
>
> Could you build such a device out of 7400 series TTL?
> If so, you can figure out how to do it in an FPGA, but much
> faster (in your time) and cheaper. &#4294967295;The thought process is
> very different from C coding.
>
> Is this a commercial product or a class project?
>
> -- glen

Did you take a look at this document?

http://www.xilinx.com/support/documentation/ip_documentation/xfft_ds260.pdf

To achieve the speed you need you may need a much larger part or one
with
more block RAM and multiplier resources.

me_rythm wrote:

> I have made a program which calculate FFT of two images (1024X1024) and
> then convolve them. I wish to know which FPGA chips would be best for this
> task, I want the result of the algorithm in real time (more than 30hz). It
> takes around 10 seconds on my machine(AMD 1.8 Ghz Dual Core) for just one
> convolution of both the images. I am thinking to use xilinx XC3S500E (500k
> gates). I have no experience in FPGA, if anyone can help me i would really
> appreciate. 

You would do better to post in comp.arch.fpga.  (added)

How many bits per sample?  Are you using the SSE feature
of the AMD machine?  That lets you process multiple 8 bit
values in one instruction, but might require assembly coding.

I suggest looking at the systolic array architecture as
it works well for many problems like this.  You don't really
need FPGA experience, but digital logic experience helps a lot.

Could you build such a device out of 7400 series TTL?
If so, you can figure out how to do it in an FPGA, but much
faster (in your time) and cheaper.  The thought process is
very different from C coding.

Is this a commercial product or a class project?

-- glen