NVIDIA's mainstream attack - 8600, 8500, 8400, 8300
Author: Luka Rakamaric
Date: 17 Apr 2007

DirectX 10 API (Application Programming Interface) has been "the next thing around in the world of graphics cards" for quite awhile now, mostly on paper and in behind-the-scenes talks. If you analyze the support in games and applications in practical terms, we're yet to see what it's all about. Up 'til now, the only VGA cards that supported the new API were quite expensive 8800 series cards, which an average computer user just cannot afford. Today NVIDIA launches the 8600 and 8500 series, basically represented by three models, the 8600GTS, 8600GT and 8500GT. Also, today NVIDIA introduces the new 8400GS/8300GS cards that are ment to be used in OEM machines.


First off, lets recap what has the 8 series brought to the table. The main thing about the launch of the 8800 series a few months ago was the unified shader architecture. The idea was to use the same processors to perform pixel, vertex, geometry and physics, rather than to have a pipeline through which the data traveled before being ready to appear on the screen. This approach allows the card to have a much higher efficiency than the previous generations, because every processor can be loaded with tasks all the time. This reduces the time that for example the vertex shader is idle because the scene is pixel processing intensive and requires much more pixel processing power. The second feature of the series 8 cards is the support for DirectX 10 API and Shader Model 4.0. DirectX 10 games will start to appear this year, and the list includes titles such as Unreal Tournament 2007, Crysis, Lord Of The Rings: Online and many more. Microsoft and NVIDIA have been working on DX10 for a few years now, and they claim that it is completely new product, quite different than what we were used to since the start of DirectX up to DirectX 9.0c. It offers geometry shaders, which allow for real time construction and destruction of objects, texture arrays and render target arrays. This basically means a completely new approach game engine creation, which is where NVIDIA?s ?The way it?s meant to be played? program comes to play. It offers developers improved support from NVIDIA, and ensures that the engine will be able to fully use NVIDIA?s cards potential.

Let?s take a look at the physical side of 8600 series cards, for starters. Both models are based on a same chip, called G84. It has 32 individual Stream Processors, but it would be wrong to just call it a fourth of a G80, because NVIDIA has made a few tweaks to the design. In series 8800 each processor can calculate four texture addresses and perform eight filtering operations. In series 8600, each stream processor can do eight texture address calculations and eight filtering operations, which allows a greater number of texture locations to be sampled. The G84 is produced by TSMC in an 80 nanometer production process and has 289 million transistors. The 80 nm process allows for very high operating frequencies. The faster card, 8600GTS, has a reference core clock of 675 MHz, a full 100 MHz more than the 8800GTX for example. Its shaders are clocked at 1.45 GHz, which is also very impressive. What the 80 nm process could bring to the table is extreme overclocking. Without voltmodding and using special coolers with Peltier effect or watercooling, it is possible to do around 740 MHz on our tests, depending on the card. Voltmodded cards have been pushed to around 1000 MHz of core clock, which is an impressive result, but severely limits the chip?s life expectancy and requires special cooling that is not usually commercially available.