The nVidia GeForce 8600GTS (aka 'G84' core) is built on TSMC's 80nm
manufacturing process and contains 289 million transistors. According to nVIDIA
the GeForce 8600GTS core is suppose to have a maximum power draw of 71W which is
a hair below the 75W the PCI Express x16 bus can provide. To be on the safe side
the GeForce 8600GTS has a six pin PCI Express power connector in extreme cases
where the videocard needs more power (like when overclocking).
By default the nVidia GeForce 8600GTS has a
default core speed of 675 MHz (MSI's card is clocked at 700 MHz) while the
stream processors hum along at 1.45 GHz. The nVidia G84 core only comes equipped
with 32 stream processors as opposed to 96/128 with the nVidia G80 core.
The memory runs at a flat 2 GHz (2.1 GHz on the MSI card), yet with the
GeForce 8600GTS the memory bus width is only 128 bits. That alone should tell
potential users that the card will be mainly memory bandwidth limited,
especially so at higher resolutions or with AA enabled. Maximum memory size is
supposed to be 256MB for all G84 based videocards.
Gone are the hard coded Vertex and pixel shaders in the G84 core, they have
been replaced with a more flexible Stream Processor (or unified shader) that
calculates both types of data. The Stream Processors run at a blistering 1.45
GHz. Traditional core clock speeds as we know it are dead, as several internal
processors are running at different speeds.
It's important to note that the GeForce 8600GT only has 32 stream processors
while the sightly better GeForce 8800GTS model has 96 and the 8800GTX 128. If
you remember, 16 stream processors equals one thread processor so there is only
two in the nVidia G84 core.
Each Thread Processor has two groups of eight SP, and each group talks to an
exclusive texture address filter unit as well as well as being connected to the
shared L1 cache. When more memory is needed, the Thread Processor connects to
the crossbar memory controller. nVIDIA's crossbar memory controller is broken up
into two 64 bit chunks for a total bus width of 128 bits.
By moving the GPU towards a threaded design, the nvidia G84 is much more like
a processor than any graphics core of the past. Any type of data - be it pixel,
vertex, or geometry shader can be processed within the Stream Processor. This
allows load balancing to occur between the various tasks.