0.18 micron manufacturing process
183MHz core clock
up to 200MHz memory clock
2 pixels per clock cycle
3 textures per pixel pipeline
366Mpixels/s fill rate
1.1Gtexel/s fill rate
approx. 1.3Gtexels/s effective fill rate w/HyperZ technology
6.4GB/s memory bandwidth w/200MHz DDR
3.2GB/s memory bandwidth w/200MHz SDR
16MB-128MB frame buffer
AGP 1X/2X/4X support
30 million triangles/sec
Transform, clipping, and lighting
4-matrix vertex blending
Full scene anti-aliasing
Motion blur and depth of field effects
32-bit color rendering
Spherical, Dual-paraboloid, and Cube environment mapping
Emboss, dot product, and environment mapped bump mapping
2K x 2K texture support
DirectX texture compression
Integrated MPEG-2 decode including iDCT and motion compensation
Support for all DTV/HDTV formats
4-tap horizontal and vertical filtering
Per-pixel adaptive deinterlacing
Optional video input/output
2-bit VIP host port
VIP 2.0 8-bit capture port
Optional DVI interface with integrated TMDS transmitter
Full DirectX and OpenGL support
Radeon cards will be available in 32MB and 64MB configurations. Currently, 32MB DDR and 64MB DDR cards are shipping and will retail for $279 and $399 respectively. A 32MB SDR card will ship in September with a $199 price tag.
The video in/out option will be available for both 32MB and 64MB cards. The Radeon uses ATI's famous Rage Theater chip for video encoding and decoding. DVI support is an optional feature for OEM cards, but will not be available in retail versions. Read our Radeon Preview for more information about the card's video features.
Charisma and fill rates
ATI refers to the Radeon's dedicated geometry unit as the Charisma Engine. The Charisma Engine can handle up to 30 million triangles per second and supports advanced character animation features such as 4-matrix skinning and keyframe animation. Check out our GDC article
for more information on these advanced animation features.
As you can see, ATI has lowered the Radeon's clock speeds since the April announcement. The Radeon supports memory speeds up to 200MHz, but shipping boards will most likely sport 175MHz or 183MHz. We aren't very concerned since increasing core and memory speeds doesn't seem to do much besides allowing the marketing department claim a higher fill rate.
ATI's Pixel Tapestry Architecture features a dual-pixel pipeline that can process three textures per pixel. NVIDIA's GeForce 2 GTS has four pixel pipelines, but can only process two textures per pixel. The original GeForce 256 also has four pixel pipelines, but can only process one texture per pixel.
The Radeon will be most effective in game settings which require three passes. This, however, does not change the fact that the GTS has a higher texel fill rate, but ATI has developed its HyperZ technology to increase the Radeon's effective fill rate by "over 20%." Even though the Radeon's theoretical fill rate is near the 1.1Gtexel/s mark, HyperZ efficiencies push the effective fill rate above 1.3Gtexels/s.