» Performance of the Card reflects the MArchitecture In a press release shortly after release of the HD 2900 XT Nvidia released a comparison of the AMD and NVidia Micro Architectures. After looking at this the performance in benchmarks falls exactly where it should. The biggest deviation is actually 3DMark06.
AMD chose the marketing teams way of representing the shader processing in the card. They say they have 320 shaders, but actually they have 64 5 way shaders. 64 x 5 = 320. The 8800 GTS is represented as having 96 shaders. NVidia in there comparison basically stated that these are 2 way shaders. 96 x 2 = 192.
This is still a lot less than the AMD cards but wait, there's more. AMD cards have a single clock at 740 MHz. 740 x 320 = 236800. NVidia has two clock domains in their chip. The main clock domain (500 MHz) and the Shader Clock Domain (1200 MHz). 192 x 1200 = 230400. This Places the GTS slightly behind the HD in theoretical shader performance. However that is not all.
AMD has 16 Texture units while NVidia has 24. At the main clock rates this puts the AMD card slightly behind the NVidia Card.
Finally there is memory bandwidth. AMD hands down has a lot more. This seems to be bottlenecked by the performance of the rest of the card. By theoretical number crunching the XT and GTS should have extremely similar performance except at extremely high resolutions where memory bandwidth makes a difference. For a clear look at how memory bandwidth effects performance compare benchmarks of the Radeon X1900XTX and X1950XTX. More memory bandwidth does improve performance, but until the processor catches up, it is only helpful for marketing, or extremely high resolution.
The appropriate price point for this card should be slightly lower than the 8800 GTS to make up for the increased energy bill, and shortened life of the rest or your computer from heat.
Even after driver enhancements this card should still theoretically be trading performance in benchmarks with the 8800 GTS. For this architecture to be comparable to the second from top of the line 8800 GTX the number of shaders will need to be widened past 420 or the clock will need to be increased past 1 GHz.
If the rumor mills are accurate then we don't have anything exciting to look forward to in the mid range products from AMD. They will fall behind there NVidia counterparts in anything requiring significant texture processing, and might pull slightly ahead in games that have extremely heavy shader processing.
I don't remember where I saw NVidia's rebuttal to the AMD architecture so if anyone knows where it is please reply with a link.
