: News
: Reviews
: Editorials

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

GeForce 8800 Technology Preview

Massively General

by Josh Walrath

Architectural Expectations

            So what exactly is the fuss all about?  I think a large portion of the excitement around this architecture is that it is far from what most were expecting.  NVIDIA has a interesting history of approaching graphics technology.  In the beginning they were very aggressive with adding features and doing their best to redefine graphics generation after generation.  From the Riva 128 to the TnT to the original GeForce, NVIDIA had pushed the technology farther than any other company in the industry at this time.  After the original GeForce the features were not pushed as fast as they were in the past.  The GeForce 2 was a shrunk and optimized version of the GeForce with a second texturing unit per pixel pipeline.  The GeForce 3 added integer pixel shading functionality, and the GeForce 4 was a further refresh of that architecture.

            After that NVIDIA suffered its first real setback with the GeForce FX series of chips.  While the debate STILL rages among enthusiasts about the merits and pitfalls of the architecture, it is hard to debate the fact that ATI had the more forward looking architecture for DirectX 9.  NVIDIA was quick to jump on the situation, and a year and a half after the disastrous introduction of the GeForce FX 5800, the NV40 series of products were introduced.  NVIDIA at this point appeared to decide that pushing performance and features was the bread and butter of the company.  Full SM 3.0 support as well as competitive performance to ATI’s X800 series made the GeForce 6 series of cards a favorite among consumers.  NVIDIA then refined the architecture with two series of G7x parts, culminating with the GeForce 7900 series based on TSMC’s 90 nm process.  At this point many thought NVIDIA had gotten conservative again with their designs.  While the 7 series was faster and more efficient than the 6 series, it was not a radical redesign and really added few features.  Transparency AA was a highly touted feature, but with the latest Forceware Drivers we see that it was a feature that was present but not uncovered in the original 6 series of products.

            This idea of conservatism was further reinforced by David Kirk (Chief Scientist at NVIDIA) when he expressed his thoughts of what the next generation of NVIDIA products supporting DirectX 10 would be like.  In these interviews and speeches David re-iterated several times that a fully unified pipeline was not the most efficient at this point in time.  From these discussions most people expected NVIDIA to embrace an architecture that still divided pixel and vertex/geometry processes into different elements in the chip.  The debates that this revelation sparked were tremendous.  Those in the ATI side of the argument pointed to the Xenos processor powering the Xbox 360 and its unified architecture and how ATI would be releasing a second generation unified architecture with its R600.  Those defending the NVIDIA side pointed to the potential transistor costs of implementing a fully unified design and how those transistors could be used to provide better acceleration for DX9 content and the first generation of DX10 applications.

            Imagine my surprise when I attended the initial G80 introduction and discovered that NVIDIA was going to go with a fully unified design.  Not only that, but the design was far wider, more complex, and faster clocked than anyone was expecting.

Next: The G80 Architecture

Copyright 1999-2006 PenStar Systems, LLC.