FX Delays/Pipelines

0.13-micron Delay

NVIDIA’s GeForce FX is the company’s first foray into TSMC’s 0.13-micron process, which is often fingered as the source of GeForce FX’s numerous delays. It’s no so much the use of 0.13-micron (with copper interconnects) that’s the culprit; rather it’s believed that TSMC’s difficulties with low-k dielectric are the primary cause. Low-k dielectric is a material used to shield the wiring within a chip, reducing electrical crosstalk.

As more and more circuits are packed tighter within chips like GeForce FX, these circuits start to generate interference in one another. Low-k dielectric prevents this from occurring, allowing electronic signals to move cleaner and faster. When GeForce FX was originally conceived it’s widely rumored that it was intended to be built with TSMC’s 0.13-micron process with low-k dielectric. But when TSMC couldn’t get good yields out of its low-k dielectric process, NVIDIA had to scrap the idea. With GeForce FX’s 125 million transistors (versus 63 million in GeForce 4 and 110 million in RADEON 9700), the chip is the most complicated design to come out of TSMC. Complicating matters even further is the chip’s 500MHz core clock frequency, which is also higher than any other graphics processor to date.

It is widely believed that NVIDIA was counting on a 0.13-micron design with low-k dielectric to make GeForce FX’s breakthrough clock frequencies possible. When low-k dielectric wasn’t ready, NVIDIA pressed ahead without it. As a result, NVIDIA has had a hard time getting good yields out of GeForce FX 5800 Ultra parts, which explains why NVIDIA has had such a hard time supplying the retail and OEM segments with cards. Unless TSMC is able to implement low-k dielectric or improve its existing 0.13-micron process, GeForce FX 5800 Ultra may never make it beyond limited availability.

8-Pixel Pipeline or 4?

Another more recent source of controversy surrounding GeForce FX has been the number of pixel pipelines it contains. At launch NVIDIA was proud to boast GeForce FX’s “8 pixels/clock rendering pipeline”. Everyone assumed NVIDIA was referring to an 8-pixel pipeline with 1 texture unit per pipe configuration similar to what ATI has employed in its RADEON 9700 architecture. However, when GeForce FX cards made their way into reviewers hands it was discovered that GeForce FX performed similarly to the 4-pixel pipeline architecture used in GeForce4 in some situations.

Everything really came into the spotlight when NVIDIA voiced its displeasure with the game tests used in FutureMark’s 3DMark 03 synthetic benchmark, specifically game test 1. To quote an NVIDIA whitepaper on the topic:

“For any given frame in the scene, up to 90% of the pixels in the frame are single textured. This occurs because the majority of the scene is the low poly-count, single textured “skybox”, painted to look like sky and clouds. The documentation for 3DMark03 gives some detail about the four-layer multitexture used on the airplanes. Regrettably, all this effort is lost on the fact that the airplanes cover so few pixels on the screen to make these four layers of multitexture completely insignificant. Game Test 1 is, essentially, a single texture fill rate test. No modern games, even DX7 games, are completely dominated by this kind of simple rendering technique.”