I don't think David was saying that only one model number can be handled per manufacturer - many user posts on many message boards would put the lie to that.
He said that only one 'type' of GPU can be handled, which in turn I take to mean that only one <main_program/> can be specified per GPU brand. Provided that the same application can run on all brand sub-types (model numbers) - just as the SETI Fermi application can run on all previously supported NVidia models with sufficient memory - then there is no problem running multiple, disparate, GPUs The problem arises if the various brand GPU sub-types require different coding, and that coding can't be combined into a single 'fat' binary supporting all cards. If that can't be done, an alternative work-round would be the chooser-despatcher model used until recently by Einstein - though that has its own drawbacks, as recently delineated by Bernd. > On 14/12/10 22:19, David Anderson wrote: >> BOINC can currently handle only 1 type of GPU per brand per host; >> i.e. it can handle 1 type of NVIDIA GPU and 1 type of ATI GPU. >> >> Removing this restriction if possible but would be a >> medium-large development project. > > Dang! > > So, does that mean that when I upgrade from an nVidia 250 -> an nVidia > 450, the 250 cannot be used by Boinc even if I leave it in place in that > host? > > :-( > > > ... Could I run two instances of Boinc on that host and restrict one > instance to one GPU and the second instance to the other GPU? > > :-) > > > Happy crunchin', > Martin _______________________________________________ boinc_dev mailing list [email protected] http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
