The Radeon X800
7 months after the release of the X800s, ATI is back with a new line of high end graphic chips. The X800’s release in May included two versions, the Radeon X800 XT Platinum Edition and the X800 Pro. Later two new cards were added, the Radeon X800 XT and Radeon X800 SE. In the end the X800 line was comprised of the following graphic cards:
All of these chips are manufactured by TSMC and have a 130nm fabrication process. The code name for the AGP version is R420. The R423, the PCI express version of the R420, comes in three versions, the XT, PE, and XT.
The X800’s architecture is an evolution of the Radeon 9700’s, which was introduced in 2002. The fabrication process is reduced from 150 nm to 130 nm, the number of pixel pipelines increases from 8 to 16 (12 for the X800 Pro) and the number of vertex engines changes from 4 to 6.
Two new functions have been added to the Radeon 9700/9800. First of all, there is the Shader Model 2.0b, a new shader 2.0 version, which makes it possible to process longer shaders (512 instructions instead of the previous 96) with ATI’s VPU. It’s used in Far Cry with the SM3.0 and SM2.0B paths. The X800, however, isn’t compatible with dynamic branching, because of its architecture and this remains exclusive to the GeForce 6800 and Shader Model 3.0.
The other function called 3Dc, is actually a compression format of “normal maps” used for bump mapping. More efficient than standard DXTC, it’s similar (1 for 2) and has improved quality compressed normal maps close to non-compressed normal maps. In practice, a game could benefit from the 3Dc for performances (thanks to a wider bandwidth) or improved quality. A 3Dc compressed normal map of equivalent size logically provides more detail than another non compressed normal map. Here again, only the future and developers will determine if this function is a plus for the X800.
R480 and R430
ATI has now released two new chips, the R480 and R430. Both PCI Express compatible, they are based on R423 architecture.
The 130 nm R480 aims is to reach higher frequencies. The R430 is a 110 nm version still manufactured by TSMC. The new 110 fabrication process of TSMC isn’t able to reach frequencies as high as the 130 nm. The advantage is only reduced production costs thanks to a smaller die. Here are the characteristics of the X850 based on the R480 and the new X800 based on the R430.
You may have noticed that they are similar to current high-end graphic cards. The X850 XT PE is only 3.8% faster than a X800 XT PE for pixel and vertex processing thanks to a 3.6% wider bandwidth. The difference between the X850 XT and X800 XT isn’t extraordinary either, 4% for pixel and vertex processing and an 8% wider bandwidth. The difference isn’t great, because compared to the X800 XT PE, it has a smaller memory bandwidth despite an identical price.
In fact, the major innovation comes with the X850 Pro. Compared to the X800 Pro, which barely reaches GeForce 6800 GT performances, it’s 9.5% faster for vertex processing and has a 20% wider bandwidth. The recommended price is equivalent. The X800 XL should be an interesting chip, similar to the X850 Pro and $50 less expensive. Only the geometrical power is slightly behind, but as this only rarely reduces performance, it shouldn’t be a major issue. The X800 should be much better than the X800 SE at an equivalent price.
These new chips are only PCI Express compatible. As ATI doesn’t release their graphic cards in AGP version (unlike NVIDIA with the 6600 and the HSI), it seems they are not putting out any new AGP products for the moment. This isn’t that dramatic as the previous X800s still have AGP versions, but the next chip generation expected out in 2005 probably won’t be available in AGP.
