The PC graphics business is famous--some would say notorious--for what are known as "kicker" products. Because of the ever-increasing complexity of bringing a new GPU architecture to market, both ATI and Nvidia develop many derivative parts from core architectures to get the most out of them. This exercise stems from two motivations: The first is to "use the whole cow" as Nvidia CEO Jen Hsun Huang is fond of saying, meaning they do more yield-bin sorting to squeeze every bit of clock out of GPUs rolling off the fab line. The other is to address microsegments of the market, especially at the mid and low ends of the market where most of the business actually gets done.
Both ATI's R4xx and Nvidia's NV4x architectures have proven their mettle in the benchmark arena, and we've seen a gaggle of derivative parts based on these two designs. We recently saw both companies make a serious play at the low end: ATI with its HyperMemory and Nvidia with its TurboCache--both attempts to swipe some market share from the PC graphics volume leader, Intel.
As both architectures move toward the autumn of their product cycles, it becomes increasingly hard to squeeze more kickers out of these core designs. But whereas HyperMemory and TurboCache, with their reduced frame buffers, seek to do more with less, ATI is taking its high midrange X800XL GPU and looking to do more with more. 256MB more, to be precise. You'll soon be seeing X800XL-based cards equipped with 512MB frame buffers. Is more memory the answer to our problems, especially since you don't get any more memory bandwidth, just a bigger data "parking lot?" Continued...