As this is our first look at the Radeon 9700 Pro, we’ll give a brief overview of the impressive technology that makes this card tick.
The Radeon 9700 Pro is powered by ATi’s R300 graphics processor. The R300 is currently being fabricated on a 150nm process, and weighs in at a whopping -- hold onto your seats -- 110 million transistors. As a comparison, Intel’s Northwood processor is roughly half that size, at 55 million transistors. That figure becomes even more impressive when we consider that roughly 26 million of Northwood’s transistors are tied up in L2 cache, so it has, in fact, less than 30 million transistors composing its core logic, while all 110 million of the R300s transistors are devoted logic (as opposed to memory cells).
With such a high transistor count, the R300 core requires a very large number of power pins, in addition to more pins to carry logic signals. The R300 is packaged in an FC-BGA (Flip-Chip Ball Grid Array) package featuring over 1000 pins. The Flip-Chip package provides more favorable thermal and electrical characteristics, allowing easier connection of multiple pins, and facilitating easier power dissipation.
A large percentage of those 110 million logic transistors are devoted to the R300s fully DirectX 9.0-compliant rendering pipeline. The DirectX 9.0 specification calls for a fully Floating Point pipeline with a minimum of 128-bit precision. Contrast that with the DirectX 8.1 specification, which calls for an Integer pipeline with only 32-bit precision, and you can easily appreciate where the R300’s 110 million transistors come into play. As game and computer graphics become more realistic and cinematic in complexity, the importance of higher precision processing becomes very important. Essentially, the difference between a scene rendered with 32-bit precision versus one rendered with 128-bit would be not totally unlike the difference between a photo taken at 100dpi versus 200dpi, for example. The former simply lacks the accuracy to place tiny pixels at sufficiently precise positions.
The R300 core features no less than 8 pixel rendering pipelines, with one texture unit per pipeline. Also in compliance with DX9, the R300 features 16 texture inputs, giving it the ability to apply 16 textures per pass (the GeForce 4 and Radeon 8500, for example, can apply 6 textures per pass). To give an idea of the massive headroom afforded by a 16 texture per pass design, most current games apply no more than 4, or perhaps 5 textures per pass. Even upcoming games are likely to require only 6 texture layers. So the ability to apply 16 textures per pass won’t translate into an immediate performance advantage for the R300, but it does afford it a considerable degree of headroom for the future.
Discuss This Article (1 Comments) - If you have any questions, comments or suggestions about the article and/or its contents please leave your comments here and we'll do our best to address any concerns.
Rate This Product - If you have first hand experience with this product and would like to share your experience with others please leave your comments here.