Is the concept of milkyway_gpu still considered?
http://milkyway.cs.rpi.edu/milkyway_gpu/


Wouldn't be better if milkyway is all in a single project (even if someday you will give GPU different work units)?

And what about the new app that was able to do work also on single precision devices and that Cluster Physik ported to CAL but was never used?

Wouldn't be better if milkyway is all in a single project (even if someday you will give GPU different work units)?

I think we'll probably making the change to do something along these lines. The real problem here is doing it in a way that doesn't make the CPUs obsolete ;)

And what about the new app that was able to do work also on single precision devices and that Cluster Physik ported to CAL but was never used?

I'm not sure if we'll ever be able to come up with a way to make single precision workunits worthwhile for either you or us. I have had a couple ideas but they require a lot of rewriting of the server code, and I'm not quite sure how much benefit it will be considering all new GPUs can handle double precision math.

---

all new GPUs can handle double precision math.

radeon 57xx can't, new 40 nm GT 2xx can't (as far, as i know).

and i think the number of users who have 57xx much more when those who have 58xx.

All the GTX models 260 up can handle double precision. The GTS models can't.

---