Good article. It suggests CPUs are heading towards parallel programming in a big way. I've been banging on about the direction on computing for a few years now; basically saying more and more cores will be used, while clocks will not improve significantly. The article explains how these will be exploited generally. Presently most CPU Boinc projects run several separate (independent) tasks simultaneously (matched to core/thread count) but some projects could benefit enormously from parallelism. To some extent the present situation is DIY parallel processing using serial programming with the scientists collating the results. Parallel programming might call for more ordered projects, but the scientists would also benefit from a more automated system, as is the case with mature GPU projects.
Core counts are continuing to rise, but it will be when the frequencies start to significantly drop that there might be a natural drive to parallel programming. This has already begun, and is most noticeable in well designed IBM systems; designed for purpose, efficiency, and with ergonomics in mind, in modern servers, with high core count but low frequency and hence lower power consumption, and in laptops. While leading desktop computers have a need for speed, this is often driven by gaming, which has diversified to the extent that only a few are prepared to pay for top specs, and those that are know a top GPU is more important than a top CPU. Few games actually need top CPUs, and in the near future some may very well require many cores to support high GPU specs; gaming programmers will be among the first to use such parallel programming techniques.
Hopefully many Boinc projects will follow, and I do think bandwidth would be slightly reduced, relative to work done, but reducing overhead.

FAH (Folding at Home), not to be confused with FAAH (FightAIDS@Home), are not compatible. Have no idea what BOINC agent porting discussion have ever taken place over at FAH.

As for running BOINC and the FAH agent on the same machine, simultaneous, think they can run same time, but not at all aware how the OS and agents coexist to have the science processes not occupy the same cores.

GPU-based crunching discussions needed a home. There has been a ton of discussions about GPU-based crunching, which would have used the GPU forum, instead of being scattered throughout other forum headings. I promise to not pressure WCG to make a GPU-aware WU (frankly though, I'm not sure if that would be a good thing for WCG ). So, how about now for that GPU forum? Please?
;

Seen on the forums just the one project having long talked about getting a GPU implementation. As what bbover3 wrote then, creating a GPU forum before a GPU launch would send the signal as were there such a project, when there is not, but making one with a sticky header post would work nicely as a honeypot. Till then, if folk can't type GPU or NVidia or GPU+NVidia or ATI+GPU to refine what's wanted in the search or advanced search box, well, that's reality of today.

:)

I would like to point out that I'm not anti-GPU. However, having said that, I do often put forth the argument that it's not the ultimate solution to every problem and there is a lot more involved in making projects GPU compatible than many of its afficianados are willing to admit.

Yes, GPUs can put out amazing quantities of results when the stars align, but there is a very heavy investment required in making everything work. For a lot of sciences, this is more effort than potential reward. As long as people kept that in mind and didn't expect miraculous results, I'd be much more supportive of the arguments.