|Radeon 9800 Pro Technology Overview|
August 6, 2003 | Lucian Fong
The battle for the speed title between ATI and NVIDIA has diminished, but the video card king has yet to be crowned. Today's mid- and high-end video cards have enough raw power to run most modern video games with nary a stutter, but much of their potential and resources lie untapped. Though they have been gradually increasing the pixel fill rate, both ATI and NVIDIA have shifted their focus to improving efficiency and image quality, going so far as to claim that their latest video cards are capable of rendering cinema-quality graphics in real-time. That may be stretching the truth slightly, but they have written some very impressive programs that demonstrate their forward-looking features. And as these new technologies trickle down into more affordable graphics cards, game developers will use them to enhance the visuals in their games.
ATI's most advanced graphics card to date is the Radeon 9800 Pro and it is fitting that the Mac version debuted alongside the Power Mac G5, which Apple claims is the world's fastest desktop computer. In contrast to the G5 introduction, the retail Radeon 9800 Pro "press bonanza" was surprisingly low key for a flagship product release. Not since the original Radeon succeeded the Rage 128 has there been such an exponential leap in performance and features over previous graphics cards. Today we dig deeper into the heart of the Radeon 9800 Pro and attempt to explain the whiz-bang technologies that ATI engineers have spent so long developing.
The CoreThe Radeon 9800 Pro is built around a completely new and incredibly complex graphics core — codenamed R350 — that contains 110 million transistors. To put that into perspective, the IBM PowerPC 970 processor found in the Power Mac G5 has 52 million transistors. Many of the Radeon 9800 Pro's transistors are dedicated to the 128-bit floating point rendering pipeline, which, in short, allows for more precise and realistic-looking images. A 10-bit per channel digital-to-analog converter (DAC) preserves much of that precision when it is output to the screen.
The number of rendering pipelines has been increased to eight, double that of the Radeon 8500. Combined with the 380 MHz core clock speed, the Radeon 9800 Pro has triple the pixel fill rate of ATI's previous flagship graphics card on the Mac. This, and the four vertex shader pipelines have the effect of drastically increasing triangle throughput to 380 million polygons/sec, which is vital to vertex shader performance.
ATI chose to equip each rendering pipeline with it's own pixel shader and a texture unit capable of sampling 16 samples per pass (the number of clock cycles per pass is variable). Anisotropic filtering, a method for improving image quality that will be discussed later in this article, should receive a healthy speed up.
Finally, the R350 core exchanges data with the 128 MB of on-board DDR-memory through a 256-bit interface — again, twice as wide as the Radeon 8500 — via four, 64-bit memory controllers. The memory is clocked at 340 MHz (680 MHz effective), resulting in nearly 22 gigabytes/sec of bandwith, much of which is consumed when full-scene anti-aliasing and anisotropic filtering are enabled at high resolution.