Re: OGR cuda client?
Steven Nikkel <steven_nikkel-WeBCG/[email protected]>
| Newsgroups | gmane.comp.distributed-net.rc5 |
|---|---|
| Message-ID | <[email protected]> |
The hardware in a GPU is excellent for use with the RC5 algorithm but it is fairly unsuitable for OGR. Distributed.net runs purely on volunteers. Currently no-one has volunteered or worked on an OGR core. We welcome any contributions people are willing to provide. While we are eager for performance improvements, we also have to be very careful about introducing bugs. A small bug could result in large scale corruption of results and lot of work to duplicate. Additionally at this stage we also suffer from the lack of time and expertise volunteers can provide to review and test cores for correctness. For the GPU clients we are especially concerned with bugs in the cores as we continue to see a wide variety of ever-changing issues in the beta and test clients. > For some reason, they started off by doing RC5 on CUDA. There is a bug entry to do OGR, but so far there has been little progress... As with all the recent core improvements. It's pretty baffling that with two projects on the roadmap that are projected to be 15 years from complete, there seems to be quite a bit of apathy/indifference on making them end faster. > > The guy who did the SSE OGR core has huge gratitude from me. It seems like it's about time to finish off the 64-bit OGR core and get them tested and out there. Further optimization can come in a later version. The people arguing on the Stream bug to hold back release so it will support dual cards drive me nuts. > --- Steven Nikkel steven_nikkel-WeBCG/[email protected] _______________________________________________ rc5 mailing list rc5-Ra3b/QYEcJ3d140v2zMXi0fjHoOT/h/[email protected] http://lists.distributed.net/mailman/listinfo/rc5