People are doing raytracing using CUDA, and getting very good performance
results, but I think the feature-set is quite limited. I understand that the
next NVIDIA GPU architecture will support full C++ on the device side, meaning
recursion will be possible. It also is more focused toward double precision,
which is good.

They're talking about a consumer release Q1 2010, with the HPC release May 2010.

Post a reply to this message