I am going to leave this discussion with one last thought, because it is starting to feel more like an argument than a friendly, cooperative discussion.
BOINC has been one of the early adopters of GPU computing. When we began working on it, OpenCL was still under development. Yes, if we were to start over from scratch, we would probably do things differently, given the knowledge we've gained. And we probably will do some major overhauls at some point. But right now, we have limited resources and a large installed base of both projects and volunteer crunchers. So for the time being, we need to build on what we already have in place and emphasize being backward compatible wherever possible so as not to break the work of our projects and volunteers. For example, we would like to make the scheduler smarter when dealing with multiple GPUs on one computer when some are more powerful than others. But this will require a major logic rework, so it won't happen in the near future. And for many of the changes some people want, there are other people who want the opposite. For example, some projects feel that having OpenCL separately associated with each vendor is important because some hardware architectures are not as well suited to OpenCl as others. Cheers, --Charlie _______________________________________________ boinc_dev mailing list [email protected] http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
