Neither ATI nor NVIDIA managed to deliver their new high-end graphics processors code-named R420 and NV40 to the market last year. Moreover, as some sources indicate now, only in very late December 2003 both leading graphics companies managed to tape out the actual next-generation graphics chips.
Precise specifications of the upcoming graphics developments from leading powerhouses are absent at the moment, a totally extraordinary situation for the recent years. What is known about the chips code-named NVIDIA NV40 and ATI R420 is that they will have more capable and powerful Pixel Shader 3.0 pipelines and Vertex Shader 3.0 processors to correspond the forthcoming DirectX 9.1 API from Microsoft. Following the latest traditions, expect some new memory bandwidth-saving mechanisms, improved speed in “eye-candy” FSAA plus anisotropic filtering modes, much higher geometry power, calculation power and so on.
Both leading-edge graphics processors will be made using mature 0.13 micron technology and work at higher core-speeds than today’s GeForce FX 5950 Ultra or RADEON 9800 XT.
ATI Technologies and NVIDIA Corp. usually have internal target to develop new generation chips with performance two times higher compared to previous generation products.
One or both graphics companies have been evaluating 1600MHz (1.60GHz) 256Mb GDDR2 memory from Samsung Electronics since late August 2003. Peak theoretical bandwidth of 1600MHz memory on 256-bit bus would be mind-blowing 51.2GB/s.
ATI has two high-end graphics chips in its roadmap for the first half of they year: R420 and R423. The former is designed for AGP 8x, the latter is developed for PCI Express x16. Other specs of both chips are supposed to be the same. NVIDIA has only one high-end chip for the 1H of the year – the code-named NV40. It is not fully clear whether this is an AGP 8x or PEG x16 part, but even in case it is engineered for AGP 8x, NVIDIA still can address the PCI Express market with it, as the company has a special PEG x16 to AGP 8x bridge.
As transpired on The Inquirer, NVIDIA’s NV40 GPU was taped out “in the last days of December of 2003”. Based on some information from sources with presumable knowledge of the matter, ATI managed to receive the first silicon of its R420 visual processing unit a bit earlier than its Santa Clara, California-based rival, though, neither ATI’s nor NVIDIA’s graphics cards partners have received samples of the chips yet.
ATI Technologies said late last year that it utilizes low-k 0.13 process at TSMC for its next-generation graphics products, moreover, an ATI’s spokesperson reassured that the company will have “no yield problems” with its next-generation graphics technology.
Currently available information suggests that the new graphics chips may be announced at CeBIT 2004, or by CeBIT 2004 with actual graphics cards from AIB companies on display at the show.
Usually it takes graphics companies from 90 to 120 days to start commercial production of graphics processors from the initial tape-out. In case the first silicon of a 0.13 micron chips does not work correctly, it may take up to 14 weeks to tape out another one with the issue addressed.
Graphics hardware companies usually deny to comment on unreleased products.