Reference Video Cards vs Third-Party Designs: What's the Difference?

By Matthew Braga

Deciding on a graphics card is hard enough, but knowing the difference between the same model across different brands can be even harder.

Extravagance is over clocking your CPU to 7GHz and cooling it with liquid helium. Extravagance is configuring a $20,000 Mac Pro with 12-cores and all the optional upgrades. And most recently, extravagance is buying a thousand dollar GPU — one of the fastest on the market — packaged in a P90-themed gun case.
What you're looking at below is the XFX HD 5970 Black Edition, a third-party Radeon card with two 5870 GPUs welded directly to the board. There's no questioning this card is fast, but for consumers, this is where confusion sets in. While AMD and NVIDIA will design and manufacture their own GPUs, those reference builds are licensed out to other manufactures too, who then modify them as they see fit. That means buying the latest Radeon or GeForce isn't as simple as walking into the store — you need to know which flavor you want. 

 NVIDIA's GTX 460 reference design.

Research and development is expensive. It's the reason there are lots of companies that produce and manufacture graphics cards, but only two that actually design discrete GPUs and the boards that house them — AMD and NVIDIA (editor's note: we're focusing on these two companies as they dominate the discrete market). Instead, it's the reference design for the original cards that manufacturers like ASUS, BFG or Diamond will license when producing products of their own.

But while you might find five or more GeForces or Radeons that look the same, the differences can be striking; what it comes down to is a simple matter of cost. First-party cards from AMD, for example, are stock models based off the reference designs that are sent out to third-party manufacturers. What makes these cards so attractive is the extra circuitry — namely, precise control over voltage for overclockers and PC enthusiasts. But those same features can make reference cards more expensive too.

In order to cut costs, many third-party cards will remove this circuitry, limiting their overclock potential. This is fine for most users who are unlikely to modify their card, but bad news for those hoping to coax any performance gains from their new purchase. Notable exceptions include some Asus cards, like the 5870/5850, though your mileage may vary.

What's important to remember is that reference cards are exactly that — a reference. Manufacturers aren't required to adhere to the exact design from ATI and NVIDIA, and can thus take liberties with the manufacture and design of the card. This is why each of the XFX 5970's GPUs are assigned 2GB of DDR5 memory, despite the fact that the single GPU version only supports 1GB. Similar changes can occur with memory and core clock speeds, where engineers will often apply their own speed increase or overclocks to compensate for the removal for voltage modifiers.
Fudzilla's comparison of the XFX 5970 (top) and the original ATI reference card (bottom) 

These custom tweaks are the reason that few companies sell GPUs in their reference configuration, and give manufacturers a way to differentiate their product from their competitors. For companies like NVIDIA, who don't manufacture first party cards, this can make finding a reference model can be difficult.

As always, the most important thing to remember when shopping for a new GPU is research. A reference AMD or NVIDIA card isn't better or worse than a third-party variant — just different. Things like cooling and clock speed can change between brands, and comparison shopping is important to get the best performance for your cash. You should also always check the card's warranty, especially with pre-overclocked SKUs. Sites like Tom's Hardware and Maximum PC are great for resources for this — as is our own community of users.
Tell us what video cards you're running, and whether it's based on a reference design of a third-party variant!
Images via Fudzilla, XFX and NVIDIA.