They are going to post some source code soon, so have fun porting this to CUDA. This would be the first time that computers can generate realistic sounds of splashing, bubbling or othererwise disturbed water. Would be nice to have this in realtime on CUDA (not sure if this is possible).
I took a look at their paper (after watching the video, which is pretty cool). A few of the sounds could still stand to be a little more realistic, but most of it was very good (and I’m sure a bit of tweaking with the algorithm, along with some 3D audio rendering would make it incredible). Their abstract says that they include ideas for parallelization in their paper, but I didn’t see it in there (though I just skimmed over it).
In any case, it looks like it would be possible to port it to CUDA, since most of their algorithm is just linear algebra. When they post the source code, maybe we can have a friendly little forum contest again.
“Based on said source code, create a CUDA implementation that computes no longer than 5 minutes on a GTX 280 and creates the most realistic sound of POURING A PINT OF BEER. Extra points for creating a graphical representation as well. The winner gets a 5l party keg of good German beer shipped to his doorstep.” ;-)
It’s already parallel code, built on some obscure remote procedure call system, also on Boost, CBLAS and LAPACK.
With some effort, it would be portable to CUDA.
You cover the flight tickets, I’ll take you to the Oktoberfest. There are no tickets, but one can make reservations - but only for sizeable groups (companies, club members, etc…) with certain minimum expenses for food and drinks. Typically you just get in line for the tents in the morning, and usually you get a seat. Maybe not on weekends ;)