Let me explain briefly what happens in the pipeline for you to understand its importance, each pixel that is rendered on your screen goes through a pipe where it'll receive its complex color/effect etc. Each time that pixel is altered it'll pass through the pixel pipeline, one pass is one clock cycle. You can imagine going from 8 to 16 pipes can bring you a dramatic performance increase. I have to note here that these are not theoretical pipelines as rumored by some websites prior to the product announcements. They are scalable though, each pipe is available at any time in sets of 4. Where NV35 was stereotyped as 4x2/8x0 architecture, the NV40 is 16x1/32x0.
Architecture Characteristics of the GeForce 6 Series
Pixel pipelines: 16
Superscalar shader: Yes
Pixel shader operations/pixel: 8
Pixel shader operations/clock: 128
Pixel shader precision: 32 bits
Single texture pixels/clock: 16
Dual texture pixels/clock: 8
Adaptive Anisotropic Filtering: Yes
Z-stencil pixels/clock: 32
The silicon itself is of course PCI-Express ready yet the one we are talking about is running internally on AGP 16x. PCI-Express has double the bandwidth of AGP8x and another plus; it allows data to be sent in both directions.
I know AGP 16x might sound a bit weird, but I was told by an NVIDIA representative that the GPU can actually handle AGP 16x internally at the core level. The 6800 Ultra is equipped with the latest available in affordable yet really fast GDD3 memory; on the 6800 Ultra it is running at 1.2 GHz (2x600 MHz), whereas today's product, the GT, is running at 1 GHz (2x500 MHz). BTW GDDR3 is designed in a way that its temperature is much cooler. It's likely we are going to see manufacturers making 6800 cards without ramsinks.
222 Million transistors
350 MHz Core Clock
500 MHz memory Clock
256-bit GDDR3 memory
To tidy things up a bit I've made a little overview
16 pipeline GPU architecture (12 on 6800 Non Ultra).
Up-to 8x more shading performance compared to the previous generation
CineFX 3.0 engine - Yes indeed... DirectX Shaders model 3.0 for really nice visual effects
On Chip Video processor.
DDR3 memory on a 256-bit Memory interface @ 1 GHz
UltraShadow II technology - Doom III; need I say more? Very well then... 3x to 4x faster than NV35 !
High Precision Dynamic Range (HPDR) technology - we'll discuss this later in the article
128-studio precision through the entire pipeline - 32-bit Color precision; this time with nice performance.
GeForce GTX 970 SLI review We review Nvidias money maker, the GeForce GTX 970 but this time in a 2-way SLI. As such we'll be going from fast to faaaaast. In this review we'll run the standard benchmarks, but we will also have...
Palit GeForce GTX 970 Jetstream review Palit is back in the house as we review their GeForce GTX 970 Jetstream edition. The product comes factory overclocked with a boost clock of 1304 MHz. The product has a custom and very small PCB, it'...
ASUS GeForce GTX 970 Strix review We review the new ASUS GeForce GTX 970 Strix edition. A product that has been overhauled in terms of PCB and cooling. ASUS also clocks the product a little faster then reference. The new DirectCU II b...
Nvidia GeForce GTX 970 and 980 reference review A while ago Nvidia launched the GeForce GTX Titan Black which we review. We never tested it as it was supposed to be a professional series and targeted card. Nvidia's Board partners however are slowl...