Radeon X300
With high-end and mid-range markets covered by the X800 and X600, respectively, it's time to turn our attention to the value segment and the new Radeon X300. Based on the RV370 core, the X300 series is—you guessed it—also not derived from ATI's Radeon X800. Instead, the RV370 GPU is inspired by the RV380 GPU that powers the Radeon X600 line. In terms of their architectures, pixel pipelines, and shader support, the two chips are essentially identical. However, the X300's RV370 graphics chip is being fabbed on a 0.11-micron manufacturing process rather than the 0.13-micron low-k process used for RV380.

The Radeon X300: PCI Express on a budget
Source:ATI

TSMC is behind the X300's 0.11-micron fabrication process, but don't expect higher clock speeds or low-k insulators. The 0.11-micron process is just an optical shrink of existing fabrication technology that can squeeze 40% more transistors onto the same unit are than 0.13-micron tech. From a manufacturing and profitability perspective, 40% more transistors per unit of die area is a very good thing. ATI also claims that it can move to 0.11-micron without reducing clock speeds or increasing power consumption, though you wouldn't know that looking at X300 clock speeds. The Radeon X300 and X300 SE will both run with a 325MHz core clock speed that's much lower than the 0.13-micron X600s, whose cores run at speeds up to 500MHz.

In addition to sharing a common core clock speed of 325MHz, the X300s will also share a 400MHz memory clock. However, the X300 will enjoy double the memory bandwidth of the X300 SE thanks to the latter's narrow 64-bit memory bus; the vanilla X300's memory bus is 128 bits wide.

According to ATI, X300 clock speeds are low not because of any difficulty getting 0.11-micron GPUs running at higher speeds, but to reduce the cost of X300 boards. Lower memory clock speeds let board manufacturers use serial termination and lower core clocks require fewer board layers, both of which enable manufacturers to produce inexpensive cards. More conservative core clock speeds will also allow X300s to run with passive cooling.

Although the X300s will initially only be offered as OEM products, I can't help but wonder if a few of ATI's more enterprising partners will be tempted to build overclocking-friendly versions of the X300 with faster memory and more robust cooling. It just seems like a waste to saddle the world's first 0.11-micron GPU with a pokey 325MHz clock speed.

A huge gap in left field
Looking at ATI's PCI Express X-series lineup as a whole, it's hard to miss the massive gap between high-end X800s and mid-range and value X600s and X300s. For starters, there's the technology disparity. The Radeon X800s support GDDR3 memory, longer shader programs, and 3Dc normal map compression that the X600 and X300 lines can't match. There's also a huge gap in theoretical fill rate and memory bandwidth; check out the gaping hole between the Radeon X600 XT and X800 Pro:

The X800 Pro offers almost three times the fill rate and nearly two and a half times the memory bandwidth of the X600 XT. Granted, theoretical fill rates and peak memory bandwidth don't always directly correlate with real world performance, but it's easy to see a huge performance and features gap between the X600 XT and X800 Pro.

To be fair, ATI's first wave of PCI Express products are targeted at OEMs and system integrators rather than enthusiasts and gamers who build their own systems. Among OEMs, ATI's deep lineup of mid-range and value X600 and X300 products should be enough, especially with a high-end option in the X800. However, ATI really needs to fill out the middle of its lineup before PCI Express hits retail, ideally with a stripped-down version of its X800 technology.