The 9600 PRO core

As we mentioned previously, the RADEON 9600 PRO was born out of ATI’s need to produce a DirectX 9 card for the mainstream that was less expensive to manufacture. In order to accomplish this, ATI did two important things for RADEON 9600 and RADEON 9600 PRO.

First, the number of pixel pipelines was reduced from eight in RADEON 9500 PRO (the same number used in the RADEON 9700 family), to four. By reducing the number of pixel pipelines, overall transistor count decreases: while the R300 family sports 110 million transistors, RADEON 9600 PRO contains just 60 million, reducing chip complexity dramatically. This puts it more in line with GeForce4 (a 63 million transistor chip) in this regard. Adjoining each pixel pipeline is one texture unit, resulting in a 4x1 architecture.

As a result of the reduced pixel pipelines, RADEON 9600 PRO’s fill rate is reduced in comparison to RADEON 9500 PRO. ATI attempted to offset this by cranking up the clock to 400MHz (ATI’s fastest core from a clock speed perspective yet), but we’re still looking at a fill rate disadvantage of 600Mtexels/sec (2.2Gigatexels/sec in RADEON 9500 PRO versus 1.6Gigatexels/sec in RADEON 9600 PRO).

The second cost reducing change ATI has implemented in the RADEON 9600 PRO is its 0.13-micron manufacturing process (the RADEON 9500 PRO was built on TSMC’s 0.15-micron process). By moving to a smaller process, ATI is able to produce more chips per silicon wafer, assuming good yields. This reduces production costs.

Another added benefit of implementing a smaller process is reduced voltage. Consequently, the amount of heat generated by the chip is lower than if a larger process had been used. This allows ATI to use less exotic cooling on the RADEON 9600 PRO. In fact, ATI has adopted the cooling unit from the RADEON 8500 for its RADEON 9600 PRO cards. The RADEON 9600 series also forgoes the need for an external power source. Even the four pixel pipeline RADEON 9500 wasn’t able to claim this distinction.

Finally, the RADEON 9600 PRO core is paired with 128MB of 300MHz DDR memory (600MHz effective) this boosts memory bandwidth to 9.6GB/sec, an improvement of nearly 1GB/sec. ATI has optimized the RADEON 9600 PRO memory controller for use with a 128-bit interface, the 9500 PRO controller was essentially a 256-bit controller that was chopped in half. In addition, a new 8:1 Z-compression ratio increases the efficiency of ATI’s HYPERZ technology. This should yield more effective memory bandwidth to feed the 9600 PRO.