The key frame interpolation (choosing an object's placement in "key" frames and letting the hardware create the in-between frames on the fly), while saving space and bandwidth, may be a harder sell to developers because it would not be work that could immediately translate onto other hardware, a critical path for any developer that is not receiving funds directly from a manufacturer to support a new feature.
In addition, ATI has taken measures to ensure their dominance in the video market by incorporating adaptive de-interlacing and support for their Rage Theatre companion chip.
Will the Radeon 256 be enough to dominate the cutthroat accelerator market? Let's take a closer look at the chip itself to get a better idea.
By leaping headlong into the .18micron process and bypassing the larger die size, ATI has allowed for a higher level of complexity in their first generation T & L silicon. As a matter of fact, the Radeon packs over two million more transistors then Intel's Pentium III 1GHz CPU, a whopping 30 million, making it the most advanced GPU to date. Despite this level of intricacy, ATI has managed to significantly increase clock speed as compared to their current offerings.
Whereas ATI's Rage 128 Pro was essentially a glorified Rage 128, the Radeon 256 is a completely redesigned chip with the purpose of delivering performance unprecedented for the historically OEM-based Canadian company. Packing dual pixel pipelines running at 200MHz each and three texture units, the maximum texel fillrate for the chip equates to:
200MHz * 2 pipelines * 3 texture units = 1,200Mtexels/s
It is important to note that this fillrate will be realized only when three textures are being rendered per pixel.
When asked how the performance of the 2 pipline/3 texture unit specification of the Radeon will compare to other next-generation cards, ATI had this to say:
"Leading developers have stated that they will be using more and more multitexturing in future games to improve the detail and realism of 3D scenes. NVIDIA's architecture is optimized for two textures per polygon, but as soon as you add bump maps, reflections, and detail textures, its performance drops off rapidly. The main problem is that in today's graphics processors, memory bandwidth limitations overshadow fill rate limitations, so adding more pipelines doesn't help performance when the memory interface is the bottleneck. Rather than going for all-out fill rate, RADEON 256 tackles the real problem by improving memory bandwidth efficiency with its unique HyperZ technology."