: News
: Reviews
: Editorials

: Archives
: About
: Contact
: Advertising
: Privacy
: Links
: Forums




The State of 3D


Economies of 3D

April, 2006

by Josh Walrath


            At the dawn of 3D, to start a successful company all that was needed was a handful of talented engineers and several million dollars of venture capital.  Companies like 3Dfx, NVIDIA, Rendition, and others all had remarkably similar stories.  In cases like 3Dfx, three engineers got together with several million dollars of backing and were able to design a product with 2 million total transistors (1 million for the raster portion of Voodoo Graphics, and the other million for the texturing portion).  Today companies like NVIDIA have over 600 engineers working on upcoming graphics chips that are over 300 million transistors in size.  Just as in any expanding market, the economic factors can change dramatically in a short period of time, and the survivors of today have been able to adjust to the market as needed.

            Today’s 3D market is not just about offering features and performance, it is more about managing large design teams across years of work to create parts that are economical to produce, yet still match the future products of the competition.  It is a very delicate guessing game about what is needed, what will be wanted, and how it will be produced.  Neither ATI nor NVIDIA has a perfect record, and because of those misses we have the very competitive environment we see today.  To see exactly where we are we need to look past the boards on the shelves and their respective price points.  Often we are jaded to the complexities of the situation because we consistently see new batches of products hitting the shelves with high end refreshes every six to eight months.

            In this ever changing landscape, today’s strengths could become tomorrow’s weaknesses.  While one competitor may have an edge over the other at any time, things can change dramatically with each generation of products.  Both NVIDIA and ATI are very evenly matched when it comes to budgets, personnel, and software tools, but the decisions made years ago and the corporate climate that each fosters provide the differentiation between these two behemoths’ products. 

Inconstant Theories of Practice

            The working philosophies of ATI and NVIDIA seem to change from generation to generation.  It seems that each manufacturer adjusts their thinking depending on market forces.  For the past several generations though, the thinking between the two companies has often been at odds to one another.

            NVIDIA started out in something of a maverick position.  The company was constantly pushing the boundaries of technology and features, often using the very latest fabrication process and pushing 3D features farther than any other company at that time.  Their first success, the Riva 128, was designed around Direct3D and OpenGL, and while it gave up some quality features it was a fast chip.  The competing Voodoo Graphics from 3Dfx was more focused on Glide and used a Mini-GL to support OpenGL.  The Riva TnT pushed the envelope further by supporting 32 bit rendering and two separate pixel pipelines (TnT stood for Twin Texel) while 3Dfx introduced the Voodoo 2 which had one raster chip with two separate texturing engines.  Going further we see NVIDIA as the first to design an integrated Transformation and Lighting Unit (T&L) and four separate pixel pipelines with the GeForce 256 and further refined the architecture with the GeForce 2.

            The GeForce 3 series was pushing things even further, with support for programmable pixel pipelines and introducing a part with 60+ million transistors at a time where the competition was sitting at around 30 million transistors.  This was done using TSMC’s new 150 nm process at that time.  Not only did NVIDIA utilize the smaller geometries of that process, but also took advantage of TSMC’s first time use of copper interconnects.  The following GeForce 4 Ti series of chips pushed performance even further and it certainly looked as though NVIDIA would never give up the performance crown.

            Then came the hiccup.  NVIDIA bet everything on an architecture that appeared to be heavily based on the GeForce 3 and 4, yet supported floating point Shader Model 2.0 rendering.  NVIDIA pushed the feature-set of GeForce FX architecture, but for once did not push the envelope of GPU design as they previously had.  This stung the company very badly, though it could be argued that if ATI had not executed so well with its R300, the GeForce FX might not be viewed as poorly as it is now.  Still, the company did not innovate on the architecture as they probably should have.  Realizing their mistake quite early, NVIDIA focused its efforts on putting out an architecture that would not only be powerful, but introduce features far above that of the competition.  The NV4x series of chips was a massive turnaround for the company, and while it would not hold the absolute crown for single card performance, it was the first available consumer product to support Shader Model 3.0 rendering.


Next: More History


If you have found this article interesting or a great help, please donate to this site.


Copyright 1999-2006 PenStar Systems, LLC.