Nvidia GeForce3 nfiniteFX engine.
The Nvidia GeForce3 GPU nfiniteFX engine allows designers to program a virtually infinite number of special
effects and custom looks, whereas before they had to choose from a pre-set selection of effects & operations. There are
two patented Nvidia GeForce3 architectural advancements that enable nfiniteFX which are accessible through DirectX 8, these
were greatly modified by Nvidia while developing the Xbox.
GeForce4 Lightning Memory architecture & enhancements.
The new NVIDIA LMA II comprises: Crossbar Memory Controller, Z-Occlusion culling,
Lossless Z-buffer compression, Primitive cache, Vertex cache, Pixel cache, Dual Texture caches, Auto pre-charge & Fast
Z-clear.
Crossbar memory controller has enhanced load balancing algorithms for different
memory sections and improved priority system for better use of memory.
Dual Texture caches
with new algorithms than GeForce3 for better 'look ahead' for
improved 3 & 4 texture performance.
Lossless Z-buffer
the Z-buffer is an integral part of the 3D process. The Z-buffer
stores how deep the pixels will be displayed on the monitor. Overdraw is where a pixel or polygon that cannot be seen on the
screen is rendered and outputted to the monitor. Nvidia's Visibility subsystem stops this wasted rendering and is identical
to ATI's Hierarchical-Z system. Nvidia's system makes a comparison of the values in the z-buffer and will discard the values
(and associated pixels) that will not be visible on the monitor before sending the data to the frame buffer.
Same as with GeForce3. NVIDIA's lossless compression algorithm can
give as much as 4 to 1 compression with no loss of overall image quality or Z depth accuracy. This gives good reduction on
memory bandwidth use.
Z-Occlusion culling
has an improved algorithms. LMA II optimizes the read, write &
selection of pixels to be rendered giving up to 25% more efficient discarding of unseen pixels than the GeForce3.
Fast Z-clear from
the GeForce3 is a very fast algorithm that sets flaged areas of the the z-buffer values to zero instead of the whole z-buffer
with a saving in bandwidth.
Auto pre-charge
as DRAM uses capacitance to remember its state and must be refreshed
which causes delays in getting the data. Auto pre-charge is logic based to guess what rows & colums in the DRAM array
are next to be accessed. It then charges them before they are used which reduces the time the GPU need to wait if it needs
to get data from them.
This with the Quad cache help speed up the GPU. Auto pre-charge with
its effective lower latency memory access thus gives more useful memory bandwidth.
nfiniteFX II.
GeForce3
only had one vertex shader, GeForce4 now has two and they are more advanced. Both units are multi-threaded and the multi-threading
is done in the chip.
New anti-aliasing engine called Accuview lets the new 4x AA mode under Direct3D applications
called 4XS. 4XS has more texture samples per pixel than 4X to be able to create better looking AA image. Accuview makes anti-aliased
look better and work faster