NV3X Architecture Overview

NV3X Architecture Overview - Page 1

NV3X Architecture Overview
A Look At What Is To Come...

By -Chris Angelini
July 25th, 2002


ATI didn?t fail to impress us with the R300 launch. Clearly, the RADEON 9700 will be the next leading performance graphics solution for the high end, at least for a while.  However, NVIDIA is not sitting idle, watching ATI capture market share.  The release of DirectX 9 is approaching and the finishing touches are being applied to NV30, NVIDIA?s successor to the GeForce4.  If you think of R300 as Germany?s Bismarck battleship, NV3X can be likened to Britain?s HMS Hood, sent to intercept and destroy the powerful adversary.  Of course, NVIDIA hopes the outcome of the confrontation won?t end the same way.  HMS Hood met a fiery demise at the hand of Bismarck?s 15? guns.  The following is a brief overview of the NV3x architecture, based on what we know at this juncture and are allowed to disclose to you.  NVIDIA's is turning its big gun toward ATi.  The battle has just begun.


That special sauce - Shaders Version 2.0
DirectX 9 requirements and beyond

Both the R300 and NV3X architectures are similar in several ways.  The DirectX 9 standard is one of the most influential reasons for this.  To begin with, both the Pixel and Vertex Shader specifications have been updated to version 2.0.  With DirectX 8, the GeForce4 is limited to a maximum of 128 instructions per vertex shader program.  NVIDIA?s ?CineFX? architecture increases that number to 1024 static instructions, giving developers a greater level of flexibility for creating more complex effects.  Additionally, dynamic flow control allows loops, jumps and branches within the shader programs, increasing the total number of vertex instructions to a theoretical maximum of 65,536. 


Images taken from Jason Mitchell's SIGGRAPH 2002 Presentation

Pixel processing has also been elevated in importance.  DirectX 8?s Pixel Shader 1.1 specification allows for four textures per pixel (version 1.4 can do six).  Shader programs are limited to four texture instructions and eight color instructions.  NV3X, on the other hand, supports shader programs 1,024 instructions long with up to 16 textures per pixel for procedural effects that do not require texture maps.  Both ATI and NVIDIA are citing procedural wood as an example of a pixel shader 2.0 effect.  ATI?s R300 can also apply 16 textures per pixel, but is limited to 160 instructions per pass.  This is where NVIDIA stands to gain a significant advantage over ATI ? if developers choose to make heavy use of long pixel shader programs, R300 will require multiple passes to process a shader program whereas NV3X will not.

NV2X Versus NV3X - Direct X Capabilities
Click image for view

Image taken from NVIDIA's CineFX Technical Brief

Color Precision and Conclusion 

Tags:  ECT, arc, view, IE, AR, V3

Related content