Almost two years ago both ATI and NVIDIA had to start making decisions about the GPUs that they would be releasing today. ATI made the decision to build R300 on a 0.15-micron process and NVIDIA was firm in their beliefs that TSMC's 0.13-micron process would be mature enough to use for Fall 2002 release of NV30.
As with any decision, the paths ATI and NVIDIA chose to go down led them through different obstacles. ATI's hybrid ATI/ArtX design team had to deal with the inevitably large die that a ~110M transistor part would have on a 0.15-micron process. To put things into perspective, this is a chip with more transistors than Intel's upcoming Prescott core, which will be built on a 0.09-micron process and a 1MB L2 cache. ATI's research told them that they would not have a mature 0.13-micron process to manufacture R300 on so they confined themselves to a 0.15-micron process, which resulted in requirements such as a 10-layer PCB and an additional power source for the Radeon 9700 Pro.
NVIDIA enjoyed the benefits of a 0.13-micron design for NV30; they could promise higher clock speeds, more features and a smaller die size than the R300 while keeping power consumption and heat dissipation down to more manageable levels. The only risks would be if NVIDIA could not get the chip design ready in time or if TSMC's 0.13-micron yields were not high enough to produce the chip in a profitable fashion.
NVIDIA wasn't the only one that had to worry about yield; the massive size of ATI's R300 would almost guarantee that yields would be low on the chip. But when push came to shove, ATI was able to introduce their R300 on time and are a matter of days away from commencing with volume shipping of the first Radeon 9700 Pro boards. ATI did find a way to balance the yield issue by allowing their partners to produce a regular version of the Radeon 9700 with a lower core clock instead of the 325MHz clock of the Radeon 9700 Pro.
This leaves us with a very important question: where is NV30? There has been a lot of speculation about NV30 and with the chip finally taped-out we're able to shed some light on the topic of NVIDIA's next-generation GPU.