With a mainstream graphics card, like the AMD6850, one can finish a Astropulse unit in under 1 hour.

With upcoming hardware, like Nvidias 28nm Kepler scheduled for 2011, it should be possible to process one unit in under 15 minutes. That's about 1000 times faster compared to the original Astropulse runtimes a few years ago.

So I was thinking: Is it about time to re-define the Astropulse algorithm? Get rid of limitations back from the old days? Increase FFT sizes, etc.?

I am no a astrophysicist - so I don't fully understand the science behind it all. But I am a curious mind :-)

So I'd like to know which design compromises have been made a few years ago - when there was no GPU computing. And what would be made differently today - in a Astropulse 2.0 application?

Cheers,
Frizz