Radeon DDR architecture features

A new chip from ATI was announced in May 2000. During this time, motherboards based on NVIDIA GeForce2 GTS had already become considerably cheaper, so it was not so easy for new products from ATI to take their rightful place in the market.

New 3-D graphics capabilities that were implemented in the new chip:

Geometric engine Charisma Engine - hardware implementation of coordinate transformation, lighting installation, clipping (clipping of polygons not included in the final frame), vertex skinning, keyframe interpolation (interpolation between key frames), perspective devide, triangle setup
Support for up to four skinning matrices used for interpolation of polygon vertices (vertex blending)
HW TCL Performance: 30 million textured polygons per second (peak)
Hardware installation of 8 light sources for the entire scene (directional aka infinite and point Lights aka local)
Full support for OpenGL and DX7 - Tranform & Lighting, Cube environment mapping, projective textures and texture compression
Pixel Tapestry rendering engine allowing three texturing units per pixel rendering pipeline
The following texture mapping methods are supported: Cubic, Spherical and Dual-Paraboloid
Rendering at 16 and 32 bit color depth
Hardware support for bump texturing of the following types: Embosing, Dot Product3, EMBM
Textures up to 2048x2048 @ 32 bit are supported
Programmable multi-texture blending modes
3D textures are supported, this allows you to reproduce volumetric effects, such as fog or dynamically changing light sources, such as fire in a fireplace
Supports the implementation of hardware effects such as Motion Blur, Depth of Field, whole scene anti-aliasing (FSAA), etc. via D3D8
Template buffer (stensel): 8 bits
Z-buffer: 16/24/32 bits
Overlaying shadows for each individual light source; for this purpose, a special Priority Buffer is implemented in the chip
Full support for Direct3D lighting model
Table and vertex fog support

Structural diagram of the video card:

Video:
Supports hardware decoding of all HDTV formats
All ATSC resolutions supported, including 1080i
Thanks to support for the YPrPb format, it is possible to directly connect HDTV displays
Supports Adaptive de-interlacing technology - a unique technology from ATI that allows you to play video in high definition without artifacts or blur.
8-bit alpha mode for mixing video and graphics is supported (for example, this is used for overlaying subtitles or animated menus)
Fully compatible with ATI's Rage Theater companion chip.

Charisma Engine

The geometry engine Charisma Engine had the following features: Hardware support for installing vertex lighting. It is expected that when using more than one light source of the Directional Lights type (directional lighting, aka Infinite; in this case, the light source is treated as a point located at an infinite distance from visible objects in the scene), the performance drop will be less significant compared to the GeForce256. Hardware conversion of the coordinates of polygon vertices from 3D coordinates of the simulated scene to 2D coordinates of the monitor screen, taking into account distance correction, the so-called. perspective transformation operation. The hardware supports the process of cutting off polygons that are invisible in the final scene - Clipping.

Vertex Skinning

This is a method of correctly transforming the vertices of a geometric mesh in places where the model is folded, i.e. work is done only with geometry, not with textures. Textures will be drawn correctly onto correctly transformed geometry. To ensure that the geometry is transformed correctly, especially on bends and joints (for example, all joints when modeling the human body), the vertex skinning (single-skin effect) technique is used. Note that vertex skinning is a subset of the vertex blending method. Essentially, vertex blending is like alpha blending, only for vertices, not pixels (V = V1*alpha + V2*(1-alpha)). The vertex skinning method is vertex blending for vertices processed by different matrices (V = V1*M1*alpha + V2*M2*(1-alpha)). We emphasize: the correct application of textures on the folds is a consequence of the correct location of the vertices, which is achieved using vertex skinning. Vertex Skinning is used to ensure smooth, natural transitions at texture junctions, especially during movement. Vertex interpolation is done using matrices called Skinning Matrices

Keyframe Interpolation

Keyframe interpolation allows you to implement in hardware a change in the appearance of the displayed object by specifying only the start image, end image and key intermediate frames, all other transformations are performed automatically. In other words, Charisma Engine inserts the required number of interpolated frames between the “key” frames and as a result, for example, you can quite simply implement a change in facial expressions on the face of a game character.

Pixel Tapestry

In addition to the ability to hardware accelerate the geometric stage when rendering an image, RADEON, of course, was able to hardware accelerate the rasterization stage. It was for this purpose that the pixel rendering architecture and the engine with the same name - Pixel Tapestry - were created. This rendering engine is built into RADEON and was specially created to implement the work of three texture units on each of the existing rendering pipelines. RADEON has two rendering pipelines, each with three texturing units. This allows:
mix and filter up to three textures per pixel without loss of speed
implement various texture transformation methods in hardware (cubic environment mapping, projective texturing, etc.)
implement one-pass bump texture mapping using Emboss, Dot Product 3 and Environment Mapped Bump Mapping (EMBM) methods
accurately simulate the reflective properties of materials (water, metal, wood, etc.)

In addition, using the Pixel Tapestry engine, support was implemented for:
Priority Buffer, which is used to apply realistic shadow mapping from individual light sources
3D Textures (three-dimensional textures), which allows you to create complex, dynamic lighting sources, as well as volumetric fog, smoke, liquids and simplifies the handling of objects with variable geometry.

Hyper Z

HyperZ works on a tile basis, that is, based on dividing the screen into square fragments. RADEON draws the polygon first in normal order, then in tile order, and if the tile completely covers the polygon, it is discarded and excluded from further processing. This is a simple but very effective trick, since the plot of most 20D games is played out in scenes containing walls, ceilings, etc. , this approach saves up to XNUMX% of rendering time.

ATI Radeon DDR Specifications

2.5 months after the release of video cards based on NVIDIA GeForce2 GTS, ATI struck back. Its new brainchild with the strange name RADEON unceremoniously violated the sovereignty of NVIDIA products, and the GeForce2 GTS fell from the throne of the leader in 3D graphics with 32-bit color.

Quake