We're a UK Collective, doing our bit to help increase that distribution, by bringing the future that already exists to a wider audience.

Crunch, save time, save tax, save money
London Tech Jobs at Career 2.0

Java and Cuda - When Worlds Collide

mugrootbeer_m.jpg Java has been a dominate force in enterprise computing and CUDA is arguably the dominate force in scientific and financial computing on GPUs.

However after many stalled attempts the two worlds have been difficult to combine, perhaps until now.
Syracuse University has developed a Java compiler aimed at GPU computing. Phil Pratt-Szeliga, Jim Fawcett and Roy Welch, have published a paper that descibes their java to cuda compiler called Rootbeer.

Rootbeer is alleged able to take unmodified Java source (all features are supported) and then, automatically parallelize the code and launch the GPU kernel.

The java performance will depend on the problem space that it is applied to however they have published some results for appropriate applications:

Dense Matrix Multiplication: 67 times faster
Brute Force Fourier Transform: 54 times faster
Sobel Filter: 3.8 times slower (oops)

Rootbeer is available for free as open source software licensed under the GNU General Public License. And the code for rootbeer is available on GitHub https://github.com/pcpratts/rootbeer1/