Once you’ve found the GPU that best fits your needs, does it really matter which vendor sells you the card? We recently compared the prices, features, and performance of two third-party Radeon HD 4870 designs—Palit’s HD 4870 Sonic Dual Edition and Sapphire’s HD 4870 Toxic Edition—against AMD’s reference design to determine which vendor offered the best value--or if there was any difference at all.
AMD is now following Nvidia's lead by transitioning away from offering its own video cards at retail, so that it will no longer compete head-to-head with its add-in-board (AIB) partners. However, both companies will design a reference board for each new GPU and give that design to their partners as a model from which to build (they might even build the entire card and sell it to the AIBs, whose only contribution then would essentially become adding the logo to the cooling shroud). Reference designs are invariably conservative with relatively low clock speeds and less-than-premium coolers. However, these reference designs are often the only implementations available to consumers during the first few months after the introduction of a new GPU.
Third-party vendors are later given the choice of sticking with these reference designs or differentiating their products in some other way. You buy a video card for the GPU more than anything else, but third-party vendors often depart from the GPU manufacturer’s reference design by pairing it with different types and amounts of memory, by overclocking the GPU and/or the memory, by attaching a different cooler, and by offering different outputs on the mounting bracket (with HDMI and DisplayPort being the most common). Third-party vendors also often bundle software—usually a game and/or an application—with their cards. Based on those criteria, which of these three manufacturers delivers the best bang for the buck? We look at each manufacturer’s design decisions first and then compare each board’s performance head-to-head.
Ed: One of the reasons we were interested in non-reference graphics cards was because they're so rare. Back in the days of Nvidia's TNT2 Ultra, manufacturers like Guillemot (remember them?) could use especially low-latency memory and super-high core clocks to set their boards apart. Customization was far more rampant. But today's cards are incredibly complex. And with so many different GPUs and models available, most board vendors just stick to the reference PCB. Today, you can choose between bone stock cards, cards centering on the reference design, but with altered cooling (Sapphire's board represents this group here), and completely custom PCBs that try improving on the lowest common denominator with more layers, better electronic circuitry, more efficient cooling, and so on (Palit's entry, in today's story). If there's a premium to be paid for the more advanced board, is it worth it? Do you get any additional performance? These are the questions Michael will be answering.