We do. But we aren't in any rush. BOINC are on their third rewrite of
the CUDA implementation by my count. We'll let them work the kinks out
first.
Of course, if BOINC never gets it right, we will be needing a
completely original implementation.
Our eternal problem is that the BOINC interfaces aren't consistent or
stable. This often leaves us playing catch-up.
David Barnard
Synecdoche was originally forked from BOINC 6.3, which had some half-baked
GPU and coprocessor support. It was a big pain to keep up with the constant
changes to it, and we didn't really have time to keep maintaining it.
So David Barnard got rid of all the coprocessor-related messy code. And what
great idea it was. BOINC 6.4.x is making rushed and untested changes to get
the scheduler working right for GPU applications, introducing new problems
(John McLeod is warning about them). And we're free from that mess :)