The GeForce 6 series (codename NV40) is Nvidia's sixth generation of GeForce graphic processing units. Launched on April 14, 2004, the GeForce 6 family introduced PureVideo post-processing for video, SLI technology, and Shader Model 3.0 support (compliant with Microsoft DirectX 9.0c specification and OpenGL 2.0).
GeForce 6 series features
SLI
The Scalable Link Interface (SLI) allows two GeForce 6 cards of the same type to be connected in tandem. The driver software balances the workload between the cards. SLI-capability is limited to select members of the GeForce 6 family; 6500 and above. SLI is only available for cards utilizing the PCI-Express bus.
Nvidia PureVideo Technology
Nvidia PureVideo technology is the combination of a dedicated video processing core and software which decodes H.264, VC-1, WMV, and MPEG-2 videos with reduced CPU utilization.
Shader Model 3.0
Nvidia was the first to deliver Shader Model 3.0 (SM3) capability in its GPUs. SM3 extends SM2 in a number of ways: standard FP32 (32-bit floating-point) precision, dynamic branching, increased efficiency and longer shader lengths are the main additions. Shader Model 3.0 was quickly adopted by game developers because it was quite simple to convert existing shaders coded with SM 2.0/2.0A/2.0B to version 3.0, and it offered noticeable performance improvements across the entire GeForce 6 line.
Caveats
PureVideo functionality varies by model, with some models lacking WMV9 and/or H.264 acceleration.
In addition, motherboards with some VIA and SIS chipsets and an AMD Athlon XP processor seemingly have compatibility problems with the GeForce 6600 and 6800 GPUs. Problems that have been known to arise are freezing, artifacts, reboots, and other issues that make gaming and use of 3D applications almost impossible. These problems seem to happen only on Direct3D based applications and do not affect OpenGL.
Geforce 6 series comparison
Here is how the released versions of the "GeForce 6" series family compare to Nvidia's previous flagship GPU, the GeForce FX 5950 Ultra, in addition to the comparable units of ATI's newly released for the time Radeon X800 and X850 series:
(*) GeForce FX series has an Array-based Vertex Shader.
(**) AGP 6600 GT variant.
GeForce 6800 series
The first family in the GeForce 6 product-line, the 6800 series catered to the high-performance gaming market. As the very first GeForce 6 model, the 16 pixel pipeline GeForce 6800 Ultra (NV40) was 2 to 2.5 times faster than Nvidia's previous top-line product (the GeForce FX 5950 Ultra), packed four times the number of pixel pipelines, twice the number of texture units and added a much improved pixel-shader architecture. Yet, the 6800 Ultra was fabricated on the same (IBM) 130 nanometer process node as the FX 5950, and it consumed slightly less power.
Like all of Nvidia's GPUs up until 2004, initial 6800 members were designed for the AGP bus. Nvidia added support for the PCI Express (PCIe) bus in later GeForce 6 products, usually by use of an AGP-PCIe bridge chip. In the case of the 6800 GT and 6800 Ultra, Nvidia developed a variant of the NV40 chip called the NV45. The NV45 shares the same die core as the NV40, but embeds an AGP-PCIe bridge on the chip's package. (Internally, the NV45 is an AGP NV40 with added bus-translation logic, to permit interfacing with a PCIe motherboard. Externally, the NV45 is a single package with two separate silicon dies clearly visible on the top.) NV48 is a version of NV45 which supports 512MiB RAM.
The use of an AGP-PCIe bridge chip initially led to fears that natively-AGP GPUs would not be able to take advantage of the additional bandwidth offered by PCIe and would therefore be at a disadvantage relative to native PCIe chips. However, benchmarking reveals that even AGP 4× is fast enough that most contemporary games do not improve significantly in performance when switched to AGP 8×, rendering the further bandwidth increase provided by PCIe largely superfluous. Additionally, Nvidia's on-board implementations of AGP are clocked at AGP 12× or 16×, providing bandwidth comparable to PCIe for the rare situations when this bandwidth is actually necessary.
The use of a bridge chip allowed Nvidia to release a full complement of PCIe graphics cards without having to redesign them for the PCIe interface. Later, when Nvidia's GPUs were designed to use PCIe natively, the bidirectional bridge chip allowed them to be used in AGP cards. ATI, initially a critic of the bridge chip, eventually designed a similar solution (known as Rialto) for their own cards.