GeForce 8 series: Difference between revisions
No edit summary |
|||
Line 5: | Line 5: | ||
| created = November 2006 |
| created = November 2006 |
||
| entry = TBA <!--- PLEASE DON'T ADD GEFORCE 8300 HERE. Please don't speculate on rumoured products here. ---> |
| entry = TBA <!--- PLEASE DON'T ADD GEFORCE 8300 HERE. Please don't speculate on rumoured products here. ---> |
||
| midrange = GeForce 8600GT |
|||
| midrange = TBA <!--- PLEASE DON'T ADD GEFORCE 8600 HERE. Please don't speculate on rumoured products here. ---> |
|||
| highend = GeForce 8800 |
| highend = GeForce 8800 |
||
| dxversion = 10 |
| dxversion = 10 |
Revision as of 18:36, 20 March 2007
The GeForce 8 Series, or codename G80, is the eighth generation of NVIDIA's GeForce graphics cards. [1] The GeForce 8 is the third fundamentally new architecture developed by NVIDIA; the last one was GeForce 256. [2]
GeForce 8 Series Overview
The GeForce 8 series arrives with NVIDIA's first unified shader DirectX 10.0 Shader Model 4 / OpenGL 2.x architecture.[1] The design is a major paradigm shift in GPU functionality and capability. The most obvious change is the move from the separate functional units (pixel shaders, vertex shaders) within old GPUs to a homogeneous collection of universal floating point processors (called stream processors) that can perform any number of tasks. The design allows an arbitrary assignment of processing resources to maintain a more optimal workload within the GPU, because not all applications use each type of shader program in the same ratio. For instance, the GPU can allocate more resources to that task instead of pixel shading for a program that needs more vertex shader geometry power. GeForce 8 does this allocation in hardware.[1]
While GeForce 8 has a large number of stream processors, it should be noted that those processors are relatively simple compared to the shader units of older GPUs. Each is scalar and thus can operate on only one component at a time, making them simple while still being quite flexible and universal.[3] However, this simplicity results in more time needed to calculate arithmetic per processor relative to the more specialized and robust shader units within older GPUs. Previous generation shader units operate on data in a vector fashion. The simplicity of G80's processors is mostly compensated by the high clock speed at which they run. GeForce 8800 runs the various parts of its core at differing clock speeds (clock domains), similar to the operation of the previous NVIDIA G7x GPUs. The stream processors of 8800GTX, for example, operate at a 1.35 GHz clock rate while some of the chip is at 895 MHz with other intermediate rates throughout.[1][3]
NVIDIA's new chip also performs correct texture filtering, a major upgrade from the previous generations that used various optimizations and visual tricks to speed up rendering while impairing filtering quality. It correctly renders an angle-independent anisotropic filtering algorithm along with full trilinear texture filtering. NVIDIA has also introduced new polygon edge anti-aliasing methods, including the ability of the GPU's ROPs to perform both Multisample anti-aliasing (MSAA) and HDR lighting at the same time, correcting various limitations of previous generations. GeForce 8 can perform MSAA with both FP16 and FP32 texture formats. GeForce 8 supports 128-bit HDR rendering, an increase from prior cards' 64-bit support.[3] The chip's new anti-aliasing technology, called coverage sampling AA (CSAA), uses Z, color, and coverage information to determine final pixel color. This technique of color optimization allows 16X CSAA to look similar to a 16X MSAA mode.[3] The advantage is that 16x CSAA can be performed far faster than 16x MSAA, or even 8x MSAA.
Another addition is the capability for the GPU to use its processors for physics rendering, a technique named Quantum Effects Technology by NVIDIA. Additionally, NVIDIA has created the Compute Unified Device Architecture (CUDA) technology. This is an interface for the GeForce 8 cards that provides general purpose functionality for the processors within the GPU, which is a developing technique, known more generically as General Purpose GPU (GPGPU).
The new GeForce 8 series also supports 10-bit display output, up from 8-bit on previous cards. This potentially allows higher fidelity color representation on capable displays. NVIDIA's PureVideo HD video rendering technology is an improved version of the original PureVideo introduced with GeForce 6. The HD edition includes GPU-based hardware acceleration for decoding HD movie formats, post-processing of HD video for enhanced images, built-in High-bandwidth Digital Content Protection (HDCP) support at the card level.[4] The GeForce 8 series also supports the NVIDIA Scalable Link Interface (SLI) for multi-card rendering.
GeForce 8800 Series (G80)
The 8800 series was launched on November 8, 2006 with the release of the GeForce 8800GTX, with 768MiB RAM, and the less powerful 8800GTS[5], with 640MiB RAM. The 8800 series replaces the GeForce 79x0 series as NVIDIA's top performing consumer video card. GeForce 8800 GTX and GTS use identical GPU cores, but the GTS model disables parts of the GPU and reduces RAM size / bus width to lower product cost.
The 8800 GTX has 8 groups of 16 stream processors, for a total of 128 processors. 8800 GTS, in comparison, features a G80 processor with 2 of the 8 groups disabled, leaving 96 stream processors arranged as 6 groups of 16.[3] It is estimated that the GeForce 8800 GTX has ~345 GFlops performance (128 processors at 1.35GHz capable of a throughput of 1MAD per clock cycle)[6]. Both the 8800GTX and the 8800GTS are built on PCBs larger than any previous consumer graphics card, with the 8800GTX measuring 10.6 in (~26.9 cm) in length and the GTS measuring 9 in (~23 cm). This raises the concern that the graphics cards will not fit inside some smaller computer cases. Both cards have two dual-link DVI connectors and a HDTV/S-Video out connector. The 8800GTX requires 2 PCIe power inputs, to keep to the PCIe standard, while the GTS requires just 1.
Nvidia released a 320MiB version of the 8800GTS on February 12, 2007 in order to tap into a more mainstream market. Aside from the decreased amount of video memory, all other aspects of the 8800GTS remained unchanged. The unit retails at US$299. At this price range, it is expected to compete with the ATI Radeon X1950XT, in addition to Nvidia's own GeForce 7900GTX and 7950GT. Such competition will likely lower demand for the aging DirectX 9 parts, resulting in price drops. [7]
Although a minor manufacturing defect related to a resistor of improper value caused a recall of the 8800GTX models (not the 8800GTS) just two days before the product launch, the launch itself was unaffected.[8]
As of March 2007, the G80 is the largest commercial GPU ever constructed. It consists of 681 million transistors covering a 480 mm² die surface area built on a 90 nm process.[1]
Technical summary[9]
Model | Release Date | Codename | Fabrication process (nm) | Core clock max (MHz) | Fillrate max (billion texel/s) | Shaders | Memory | |||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
Stream Processors | Clock (MHz) | Bandwidth max (GB/s) | Bus type | Bus width (bit) | Megabytes | Clock (MHz) | ||||||
GeForce 8800 GTX | 8th November 2006 | G80 | 90 | 575 | 36.8 | 128 | 1350 | 86.4 | GDDR3 | 384 | 768 | 1800 |
GeForce 8800 GTS | 8th November 2006 | G80 | 90 | 500 | 24 | 96 | 1200 | 64 | GDDR3 | 320 | 640/320 | 1600 |
Future development
- According to The Inquirer, NVIDIA is preparing a faster version of the G80 for launch around the same time that ATI launches its new card. The full article can be found here.
- NVIDIA's 8 Series mid-range cards, the 8600 GTS and 8600 GT, and the lower-end cards, the 8300 GT and the 8300 GS, are expected to be released April 17th 2007; prior to the time ATI will release the Radeon R600 in 2Q 2007. The full article can be found here, [2] and [3].
- There will be no mobile version of the G80 core, as it is too power hungry for notebooks on a 90 nm process. Instead, the mobile version will have the code name G81 and will be built on an 80 nm process.[citation needed]
- According to X-bit Labs, the ForceWare 100.41 drivers include information on unreleased GeForce 8 series products. These products are the GeForce 8800 Ultra, GeForce 8600 GTS, GeForce 8600 GT, GeForce 8500 GT, GeForce 8400 GS, and the GeForce 8300 GS. ([4])
- The Inquirer has said that in April NVIDIA will release an 8 series card for the older AGP slot.
See also
References
- ^ a b c d e Sommefeldt, Rys. NVIDIA G80: Architecture and GPU Analysis, Beyond3D.com, November 8, 2006.
- ^ Q3 2007 NVIDIA Corporation Earnings Conference. Nvidia.com. November 9, 2006.
- ^ a b c d e Wasson, Scott. Nvidia's GeForce 8800 graphics processor, TechReport.com, November 8, 2006.
- ^ Shrout, Ryan. NVIDIA PureVideo HD Technology - Is the PC Ready of HD Video?, PC Perspective, December 5, 2006.
- ^ GeForce 8800 Press Release, NVIDIA.com, accessed November 9, 2006.
- ^ "NVidia CUDA forums" [1]
- ^ Shilov, Anton. Nvidia Prepares GeForce 8800 GTS 320MB, xbitlabs.com, January 10, 2007.
- ^ "Visionary". All 8800GTX Cards Being Recalled, VR-Zone.com, November 6, 2006.
- ^ GeForce 8800 specifications, NVIDIA.com, accessed November 9, 2006.
External links
- Beyond3D's G80 Architecture Diagram #666 - great diagram of how things work
- bit-tech's GeForce 8800 GTX review
- Guru 3d Review of Geforce 8
- Forceware Drivers for the GeForce 8800 series only (v. 97.92)
- NVIDIA's GeForce 8 series page
- NVIDIA GeForce 8800 GPU Architecture Overview - a somewhat longer and more detailed document about the new 8800 features
- OpenGL Extension Specifications for the G8x