I think that I know the issue of all this... Yesterday to my cousin, died your first Sapphire 7950 (11196-19-20G) and like no, SN: A1351000286XX and today I see the card and all seem be ok, capacitors are ok but in the heatsinks of VMR's i saw an oil slick (and i remember that some of my gpus dead also had this oil slick) so I suspect seriously that any VRM exploited or burned and why Hynix VRM hasn't this problem, perhaps ELPIDA is a holy crap and Sapphire not be guilty of all.
So I recommended if you have any Sapphire (11196-19-20G) with SN starting for A1350 or A1351 not upload the memory core more than 1250 Mhz and if you have money, buy an accelero xtreme 7970 your VRM heatsinks are 100x better than Sapphire.
I got the A1351.
Afaik
A135100028XXX and
A135100029XXX have been conflictives for me. These gpus at 24/7 not lasted more than 2 months mining, in differents rigs, with different hardware and configuration but all the dead gpus had one thing in common the PN reference in all cards was 299-1E249-100SA and the SN starting always at A1350 and A1351.
My last 2 dead cards:ASIC = 69,4% ---> 1,168v @ 1100 core - 1480 memory | 75%-fan-fixed | temps between 66º and 79º | SN: (A135000034XXX) (buy in France)
ASIC = 76,1% ---> 1,125v @ 1080 core - 1250 memory | 80%-fan-fixed | temps between 61º and 72º | SN: (A135100028XXX) (buy in Spain)