NVIDIA GeForce 8 technology

November 2006 NVIDIA introduced their brand new Geforce 8800 family starting with the 8800 GTX and 8800 GTS 640MB. Later on we saw mainstream cards like the 8800GTS 320MB, 8600GTS and 8500GT appearing, though it seems that NVIDIA still wasn't pleased with their Geforce 8 product range because only few weeks ago they brought out yet another Direct X 10 capable video card. Named Geforce 8400GS, NVIDIA has an entry-level video card which offers high-end features, the main difference is that the 8400GS is less powerful then other Geforce 8 family members. In the past we saw how the performance difference between high and low-end VGA cards was due to lower clocks and less pixel/vertex pipelines. With Geforce 8 video cards the story is slightly different; here is refresh of the architecture principles of the Geforce 8 family:



Previous generation VGA cards we make use of separate pixel and vertex processing pipelines, for the new cards NVIDIA went for a unified core design where stream processors where left the choice to either act as a vertex, pixel or geometry processing unit. 3D scenes could be rendered more efficient and so more performance could be gained with less effort. NVIDIA grouped their units together and so the possibility was being left adjust the GPU performance without too much trouble, NVIDIA just had to disable one or more shader groups. A good example, the Geforce 8800GTS and 8800GTX use the same kind of GPU, though where the GTX has 128 stream processors NVIDIA made the 8800GTS less powerful by disabling 32 shader units. With the Geforce 8400GS we see the same thing, its core is based on the Geforce 8 architecture though it has only 16 stream processors available.



The Geforce 8400GS comes with a G86 GPU which is created for the purpose of using it in low-end video cards. The remaining stream processors aren't just disabled here, the GPU has only 16 shading units in real life which should reduce the manufacturing costs quite some. To further reduce the price per GPU, NVIDIA designed their G86 to be manufactured based on the 65nm production process, this way more GPU's can be had from the same "wafer". For the end-user this means higher GPU frequencies at the same voltage level and/or lower electrical power usage which can results in a quieter and cooler running hardware.



In the above Geforce 8 chart you'll notice how the 8400GS is similar clocked as the 8500GT. In fact both cards are very familiar with each other, they are clocked at 450/900/400 and are based on the 16 shader unit rich G86 core, the main difference between them is that the Geforce 8400GS comes with a narrowed 64-bit memory bus which reduces memory bandwidth quite a bit. In our performance tests we’ll find out if the reduced memory bus has a large impact, but before we do that, let’s take a closer look at the different retail 8400GS based products available on the market today ->