I see over at folding@home they are testing a new core that is a streaming core (it sends data to the server every frame instead of once per wu)

I was curious if gpugrid was looking at doing a similar thing? It sounds like it has multiple advantages - all usable GPU's can be used without worrying about it being too slow (since data gets returned much quicker), don't need to worry about waiting for wu's to timeout if lost/person quit, no deadlines to worry about for slower GPU's.... There may be many more but those are the ones I know about.

Bob

I reall have no clue (Matt might have), but I would assume that it would cause an incredible server overload from 4000 computers trying to upload continuously and concurrently workunits.

There would be a lot of connections but each connection would be a small amount of data. Its the same amount of data overall from what it sounds like... just sent over multiple connections. I'm not sure what the net result for server load would be though.

Bob.

What they have is very nice, but that capability isn't sufficiently important for us to invest the time in developing it (for now).

Matt

Sorry I thought I had replied but it appears as though I didn't.

Thanks for the answers,
Bob.