New Features


256-Bit Bus

One of the most obvious differences between NV30 and NV35 is that the memory bus interface has been increased from 128-bit to 256-bits in width. NV30's bus was similar to GeForce 4 Ti's, with four 32-bit "crossbar" busses, but utilising DDR-II memory as opposed to DDR. NV35's bus utilises four 64-bit busses and is said to operate with both DDR and DDR-II memory types, but the first board implementations will utilise DDR memory.

A 256-bit bus obviously allows for a higher bandwidth, though with the capability for only 4 colour writes per cycle, and with Z-Cull, compression and clearing routines, the differences under normal rendering likely to be few. Possibly only at very high resolutions will it have any appreciable impact. Where it should make the most notable difference is with FSAA rendering - Multi-Sampling FSAA is relatively fill-rate free, but bandwidth intensive, so a large increase in bandwidth should improve AA performance considerably.

CineFX 2.0

NV35's Shader architecture is based on the same "CineFX" as the rest of the NV30 line, giving the same level of Pixel Shader and Vertex Shader programmability and advertised Shader precision. However, the "2.0" because the Floating Point shader performance is supposed to be doubled, and it's said that the integer units that existed within NV30 have been fully replaced by Floating Point units.

Intellisample HCT

Details on the exact technical changes between Intellisample on NV30 and the Intellisample HCT of NV35 are sketchy. However, "HCT" stands for "High Compression Technology" and NVIDIA state this is the "next step in performance" - presumably this means that the compression techniques employed with NV35 are more optimal than those in NV30, though how they have been improved is unknown.

UltraShadow

UltraShadow is one of the larger advancements of NV35, and it is designed to optimised Stencil Buffer rendering performance.

Geometric complexity of lighting in the forthcoming title "Abducted"

Stencil Buffered shadow generation requires each of the light sources to be analysed and a pass rendered into the stencil buffer. This process is both fill-rate and geometry intensive (as the above image illustrates).


NVIDIA UltraShadow

The basic premise of UltraShadow is that it allows developers to program a depth bounded region that will restrict the level of lighting calculation to a specified range of the Z Buffer, which corresponds to the area that the lighting source is most likely to affect. The net effect of this is that it minimises the shadow calculation to just that region, thus removing some unnecessary work calculating regions that will not be seen anyway.

Presently UltraShadow is not supported by DirectX, but is available in OpenGL via the use of an EXT extension.