ATI Rage - RAGE 128

In the continuing struggle to create the fastest and most advanced 3D accelerator, ATI came up with the RAGE 128. The chip was announced in two flavors, the RAGE 128 GL and the RAGE 128 VR. Aside from the VR chip's lower price-point, the main difference was that the former was a full 128-bit design, while the VR, still a 128-bit processor internally, used a 64-bit external memory interface.

• Magnum - A workstation board for OEMs with 32 MB SDRAM.
• Rage Fury - 32 MB SDRAM memory and same performance as the Magnum, this add-in card was targeted at PC gamers.
• Xpert 128 - 16 MB SDRAM memory and, like the others, used the RAGE 128 GL chip.
• Rage Orion - RAGE 128 GL design specifically intended for Mac OS with 16 MB SDRAM memory, OpenGL and QuickDraw 3D/RAVE support, essentially a market-specific Xpert 128. This card supported more and different video resolutions than later Mac-specific RAGE 128 designs.
• Nexus 128 - Also a Mac-specific RAGE 128 GL design, but with 32 MB of RAM, similar to the Rage Fury.
• Xclaim VR 128 - Also a Mac-specific RAGE 128 GL design with 16 MB SDRAM memory, but included video capture, video out, TV tuner support and QuickTime video acceleration.
• Xpert 2000 - RAGE 128 VR design using 64-bit memory interface.

Rage 128 was compliant to Direct3D 6 and OpenGL 1.2. It supported many features from the previous RAGE chips, such as triangle setup, DVD acceleration, and a capable VGA/GUI accelerator core. RAGE 128 added inverse discrete cosine transform (IDCT) acceleration to the DVD repertoire. It was ATI's first dual texturing renderer, in that it could output two pixels per clock (two pixel pipelines). The processor was known for its well-performing 32-bit color mode, but also its poorly dithered 16-bit mode; strangely, the RAGE 128 was not much faster in 16-bit color despite the lower bandwidth requirements. In 32-bit mode, RAGE 128 was more than a match for the RIVA TNT, and the Voodoo 3 did not support 32-bit at all. The chip was meant to compete with the NVIDIA RIVA TNT, Matrox G200 and G400, and 3dfx Voodoo 3.

ATI implemented a caching technique it called Twin Cache Architecture with Rage 128. The Rage 128 used an 8 kB buffer to store texels that were used by the 3D engine. In order to improve performance even more, ATI engineers also incorporated an 8 KB pixel cache used to write pixels back to the frame buffer.

• 8 million transistors, 0.25 micrometer fabrication
• 3D Feature Set
  • Hardware support for vertex arrays, fog and fog table support
  • Alpha blending, vertex and Z-based fog, video textures, texture lighting
  • Single clock bilinear and trilinear texture filtering and texture compositing
  • Perspective-correct mip-mapped texturing with chroma-key support
  • Vertex and Z-based reflections, shadows, spotlights, 1.00 biasing
  • Hidden surface removal using 16, 24, or 32-bit Z-buffering
  • Gouraud and specular shaded polygons
  • Line and edge anti-aliasing, bump mapping, 8-bit stencil buffer
• 250 MHz RAMDAC, AGP 2×