ATI RADEON X800
The “retribution weapon” forged in the Canadian laboratories was known under a codename of R420. Interestingly, this chip had been originally prepared to ship with 12 pipelines, but the company executives decided to add performance right before the launch and enabled four additional pipelines. The R420 is smaller and cooler and less power-hungry than the NV40. In fact, this GPU was the next step in the evolution of the R3xx architecture, never receiving support of Shader Model 3.0, unlike the NV40.
Again unlike the GeForce 6800, an embodiment of new technologies and capabilities, the RADEON X800 is less exciting from the architectural point of view (see our review). ATI adjusted the structure of the pipelines so that the chip in fact is four-pipelined, but each of the pipelines can process a group of four pixels at a time. This organization allows considerably decreasing the performance loss suffered on enabling full-screen anti-aliasing and texture filtering. They also increased the number of vertex pipelines, from four to six, and improved the vertex processors themselves somewhat.
There’s a new version of the well-known HyperZ technology. Its purpose remained the same, though. HyperZ HD is still responsible for optimizing the communication between the GPU and the memory subsystem.
As for truly new technologies, I can only name the compression algorithm now used with normal maps. It is called 3Dc and helps to increase the level of detail of the game objects without spending much resources or compromising the image quality. Besides that, the so-called Temporal Anti-aliasing was offered to the public – it improves the quality of anti-aliasing where there’s enough speed for that.
However, the major innovation came from the manufacturing field: the new 0.13-micron technological process incorporates new dielectric material, the so-called low-k dielectric. Using this process, ATI reached beyond 500MHz, thus outpacing NVIDIA, although NVIDIA used to clock its GPUs at higher frequencies than ATI. Moreover, the relative simplicity of the R420 (about 160 million transistors against 220 million in the NV40) and the thinner manufacturing process helped to keep the heat dissipation at a level with the RADEON 9800 XT and to cool the card with a simple one-slot cooling system. As a final touch to the new product, the Canadians at last decided to use high-speed memory working at 1GHz and higher – exactly the thing that the older RADEONs needed desperately.
The family of R420-based graphics cards was originally comprised of three names: RADEON X800 XT Platinum Edition, RADEON X800 PRO and RADEON X800 SE, but the last version, with eight pipelines and a 128-bit memory bus, will only appear (if it ever does) somewhere near the fall. In this summer ATI will be offering the good old RADEON 9800 in its price sector. Some sources say that this decision came as ATI found the R420 chip yield very satisfying and didn’t want to cripple normal chips to make RADEON X800 GPUs. Thus, ATI is now offering only two new products:
- RADEON X800 XT Platinum Edition: 16 pipelines, 520/1120MHz;
- RADEON X800 PRO: 12 pipelines, 475/900MHz.
Both cards use the same PCB design and are equipped with one DVI-I connector, one D-Sub connector and one Molex power plug. The memory bus has a width of 256 bits.
R420-based graphics cards were overall successful in our tests, often surpassing GeForce 6800 products, although the R420 can hardly claim to be a revolution. Once again, a simple and efficient architecture wins where a complex and universal one loses. This is especially true for the hard working situation when we use both full-screen anti-aliasing and anisotropic filtering. Well, frankly speaking, ATI has been strong in such tests before, too.
Overall, there’s parity in the market of computer 3D graphics, only the weapons have become much more perfect and powerful. It is yet unclear who’s going to win this war, if there’s going to be any winner at all. ATI has high frequencies, low power requirements, and a time-tested efficient architecture that does nicely in modern games, while NVIDIA offers the most sophisticated and universal GPU for today, with support of next-generation pixel and vertex shaders and a bunch of unique technologies. In the future, with the release of the new version of DirectX and games that use Shader Model 3.0, this equilibrium may change, though.