Mixed GPU's not recognized properly

[Sotiri]

I have two PC's, each with two different GPU's, that can only fold one of them because the client misidentifies them.

The first has a GTX 580 and a GTX 460.
The slot which the client thinks is running on the 460 is actually running on the 580 and vice versa. The 580 slot (actually running on the 460 card) continuously fails. I believe because it's actually running on the 460. Are 580 cores different?

The second has a GTX 680 and a GTX 460 and it too mixes them up. These are obviously very different cards and the slot for 680 (actually running the 460) just freezes on "starting GUI".

I've tried every combination editing the slots for the GPU, CUDA and openCL but always reverts back to mixing up the GPU's.

I've pulled the 460's out of both rigs and the 580 and 680 are shown correctly and running fine, but I would like to put them back in if there's a fix for this issue.

[bruce]

This is a known problem (see Ticket #879) and apparently it cannot be fixed other than manually adjusting the index values associated with the GPUs.

In your first example, with the GTX 580 and 460, there's no real reason to change the index values. The FAH Client identifies both as Fermi GPUs and starts the FahCore that manages all Fermi devices. Yes, the incorrect name will be associated with the GPUs and the relative speed will appear to be incorrect, but the proper results will be produced by both GPUs.

In the second example, you have a GTX 680 and a 460 (one Kepler and one Fermi) and that does need to be fixed. WUs assigned to Fermi do not necessarily work on a Kepler and vice-versa. At the present time, however, Kepler support is still being beta tested so I can't tell you how or when this is going to be resolved.