my understanding is the s1070 is made up of 4 c1060s, but do those c1060s in the server rack go through even higher QA than if you were buying a single workstation c1060?
The S1070 gpus are not the same as the C1070 - clocks are different, and they are passively cooled rather than actively cooled. There have also two versions of S1070 if I am remembering correctly - the first with a 450MHz clock, that second with a 500MHz clock. The C1070 has remained the same through its life.
The M1070 and one of the versions of the S1070 cards are probably the same, although I am not sure which one.
is there a c1070? or did you mean c1060?
there is an m1060 which allows for 2 cpu sockets, i don’t know about m1070
on the issue of QA of workstation gpus to server gpus, are they the same? is there a difference in the quality of parts?
C1060/M1060/S1070. Sorry about that.
I don’t think anyone can answer your question, well certainly not without breaking employment contracts or NDAs. My pure speculation is this:
The GT200 is a gigantic part, and I doubt yields have ever been great. Binning is undoubtedly a fact of life. It would seem unlikely to me that parts sold at different clocks come from the same bin. That doesn’t mean there is anything different about the QA process between the bins (quite the opposite), but it does mean that the higher clock, passively cooled parts may well come from a “better performing” bin, which can hit higher clocks at the same power/thermal envelope limits. I would be surprised if the design MTBF is any different between any of the professional boards, whether they are rack mount or tower mount.