Because MW can run many steps in parallel on the multiple GPU cores of a GPU. Not all BOINC projects use algorithms that can do that; for example, Rosetta@Home found that their algorithms require so many steps that must be done one at a time and in a particular order that their workunits would actually run slower on GPUs. They could, however, run several of those workunits at the same time at that speed, but only on graphics boards with enough graphics memory - for example, about 10 at once on the most expensive Tesla board with 6 GB of graphics memory. They dropped development of a GPU version after those findings.

is there any more news yet on a expected date when the gpu clien will be running?

Thank Lawrence,

keep up the good work

Datahlp

A general release GPU Beta, for all types of AMD and NVidia GPU, from entry level card to enthusiast, spanning about 4 generations from each Manufacturer, on tens of thousands of variant systems, with dozens of different drivers for each card. There could well be more than a few incompatible systems. Possibly hundreds, with dozens of different types of failure.
This is why I would Beta test for one Manufacturer at a time, on one operating system at a time, and limit each round of testing to specific card designs (CC).
One big release for all GPU types could also cause lots of problems for other tasks too, when the system hangs, or just loses enough responsiveness to make the system unusable.
Even when tasks complete successfully there will be performance considerations; the effect on other tasks, relative performance compared to other GPU's, and other projects. Crunchers don't like buying expensive high end GPU's to crunch at 30% GPU utilization, losing system responsiveness, or having a GPU project slow down CPU projects. Low end cards taking days to complete task and high end cards that struggle to upload and download fast enough, or have too many tasks queued for Boinc to handle (four GTX590's). Some cards are a hundred times as fast as others. So what techs may consider a success is not necessarily what crunchers consider a success, and catering for everybody will not be easy.

Well said! I agree 100% and that is just the tip of the iceberg. I hope WCG concentrates on developing solid documentation and support BEFORE going into a BETA. While I admire the enthusiasm of some of the posters, it is clear that many do not have the necessary skill level needed to jump into GPU processing. Some of the cards mentioned by posters are just junk unfortunately. There is a lot more to GPU crunching than just checking a box in preferences.
Oh well, at any rate the roll out is going to be “interesting”. I can hear the whining and crying already.