05-24-2015, 10:31 PM
I feel like this has been covered ad nauseam, but I'll explain it once more.
The difference between reference design cards and OEM design cards is quality and reliability.
Reference design cards are over-engineered and excel in a variety of workloads and operating conditions. They use high-quality components and generally spare little to no expense in the card's design. Therefore they are expensive to manufacture, and the profit margins are very slim. This is not a problem for the chip designer though, as they make their money by manufacturing/selling the GPU chips or licensing the chip IP / reference design schematics, not from selling complete cards.
OEMs, on the other hand, do make their money by selling cards. Because of the slim margins on reference design cards, OEMs design their own variants of the card, using the reference design as a reference (hence the name.) OEMs achieve significant cost savings by re-using previous designs, using lower-quality components, and taking shortcuts wherever possible. Rather than being over-engineered like reference design cards, they are engineered to achieve the minimum acceptable quality for narrow use-cases -- mainly gaming -- while maximizing profits.
The resulting product is a very cheap board that just happens to use the same GPU chip as the reference design card. Neither the cooling solution nor the substitute components can typically handle the stress of compute applications, resulting in overheating and/or untimely demise of the card. Typically when a GPU burns out, the GPU chip itself is fine -- it's some other component on the card that has failed, and it failed because it was of inferior quality compared to the one used on the reference design.
It is for these reasons that we advise people to never buy OEM design cards, because they are junk and you will likely just be throwing money away.
The difference between reference design cards and OEM design cards is quality and reliability.
Reference design cards are over-engineered and excel in a variety of workloads and operating conditions. They use high-quality components and generally spare little to no expense in the card's design. Therefore they are expensive to manufacture, and the profit margins are very slim. This is not a problem for the chip designer though, as they make their money by manufacturing/selling the GPU chips or licensing the chip IP / reference design schematics, not from selling complete cards.
OEMs, on the other hand, do make their money by selling cards. Because of the slim margins on reference design cards, OEMs design their own variants of the card, using the reference design as a reference (hence the name.) OEMs achieve significant cost savings by re-using previous designs, using lower-quality components, and taking shortcuts wherever possible. Rather than being over-engineered like reference design cards, they are engineered to achieve the minimum acceptable quality for narrow use-cases -- mainly gaming -- while maximizing profits.
The resulting product is a very cheap board that just happens to use the same GPU chip as the reference design card. Neither the cooling solution nor the substitute components can typically handle the stress of compute applications, resulting in overheating and/or untimely demise of the card. Typically when a GPU burns out, the GPU chip itself is fine -- it's some other component on the card that has failed, and it failed because it was of inferior quality compared to the one used on the reference design.
It is for these reasons that we advise people to never buy OEM design cards, because they are junk and you will likely just be throwing money away.