Architecture

There’s a lot more under the EVGA e-GeForce 8800 GTX’s hood than an ungodly amount of transistors that makes it one of the most important breakthroughs in gaming hardware in a long time. Let’s take a look:

NVIDIA’s GigaThread technology can perform thousands of independent, simultaneously executing threads for unprecedented scalability while maximizing GPU utilization. The 768 MB of 900MHz GDDR3 RAM with an effective clock of 1800MHz and a blistering bandwidth of 84.4GB/s, is hitched to a 384-bit memory controller. EVGA’s e-GeForce 8800 GTX isn't just a blazingly fast DirectX 10 card, but a blazingly fast DirectX 9 card as well. A single GeForce 8800 GTX can easily muscle its past any current NVIDIA or AMD/ATI Radeon video card in single or SLI and Cross Fire configurations.

What really makes the 8800 GTX special is its new Unified Shader Architecture for DirectX 10. A key element of that architecture is the 128 Stream Processors inside every GeForce 8800 GTX that can execute vertex, pixel, geometry and even physics instructions as needed.

Previous GeForce GPUs relied on separate, dedicated shader units under DirectX 9 to handle vertex and pixel processing. When game scenes are rendered, the amount of complex vertex and pixel processing involved is not monolithic but dynamic, varying from one second to the next. If a scene requires more vertex shader than pixel shader processing, the GPU's vertex shader units are maxed-out handling the load, while a large portion of the pixel shader units remain idle while processing the lighter pixel shader load. When a scene requires more pixel than vertex shader processing, the pixel shader units, the situation is reversed. In short, both shader units take turns getting overworked—and under-utilized.

Meanwhile, gamers go through all kinds of drama with beta drivers, endless exercises in overclocking, or jumping on every new “speed bump” video card upgrade to get more performance out of their games at the highest resolutions and visual effect settings. But throwing more processing power and money at the same fundamentally inefficient shader processor design, while trying to wring more performance out of games that are increasingly becoming more visually complex—is a losing game.

All of that changes with Microsoft’s DirectX 10 and NVIDIA’s Unified Shader Architecture, where all the GPU shader processors can dynamically execute and load-balance any type of shader code at a given time. This makes more efficient use of the hardware and will allow future games to be created with incredibly stunning, photo-realistic visuals currently unavailable under DirectX 9. To insure that there's little or no performance hit while this is going on, NVIDIA equipped the 8800 GTX with a core speed of 575MHz—but the stream processors race along at 1.35 GHz. Although the functionality has not yet been enabled in the Forceware driver set, NVIDIA has made the stream processors overclockable for even more performance!

Antialiasing and Visual Quality

Every new advancement in video card technology for enthusiasts has been a compromise between performance and visual quality. Antialiasing was designed to eliminate the stair-step effect (also know as "the jaggies") along the edges of objects in a game. High Dynamic Range (HDR) lighting makes the environment and objects in a game more three-dimensional and life-like, by using lighting effects that approximate what you'd see in the real world.

The compromise is when you increase the amount of antialiasing while introducing complex lighting effects to improve image quality. Each increase in antialiasing, particularly at higher resolutions, incurs a performance penalty on the game's frame rates. Before the introduction of the 8800-series cards NVIDIA GPUs were incapable of rendering games with HDR lighting and antialiasing enabled at the same time.

Radeon X1800/1900/1950-series GPUs had no such limitation thanks to the well-known "Chuck Patch" (which was eventually incorporated into the most recent CATALYST drivers). Still, there were clearly limits as to how much you could turn up antialiasing at high resolutions with HDR lighting enabled, and keep a game playable with all of its visual effects turned up as well. Even antialiasing and anisotropic filter alone aren’t always enough to improve visual quality. Alpha (transparent) textures used for rending foliage, chain link fences and similar objects can still display jagged edges on closer inspection.

One exciting development that you don't have to wait for Windows Vista and DirectX 10 to take advantage of, solves these problems—NVIDIA's Lumenex Engine. Available only on NVIDIA GeForce 8800 GTX and GTS cards, the Lumenex Engine introduces Coverage Sampling Antialiasing (CSAA)—an improved antialiasing algorithm that allows ultrahigh quality 16x multisampling antialiasing at near 4x antialiasing performance levels. Texture filtering has also been improved.

Today's games typically support HDR lighting that uses 16 bits per color component (red, blue, green and alpha) for a total of 64 bits for HDR. Before the GeForce 8800 and NVIDIA’s Lumenex Engine came along, playing The Elder Scrolls IV: Oblivion on a previous generation NVIDIA card forced you to choose between inferior “Bloom” lighting and antialiasing—or HDR lighting and a game that looked like someone took a hacksaw to it. The Lumenex Engine solves that problem with 32-bit, floating-point scalar precision for each color component. This translates into an incredible 128-bits of vector-precision HDR rendering power that is unmatched by many renderers used in film! It also allows 8800-series GeForce cards to run HDR lighting and antialiasing at the same time—a first for any NVIDIA card.

The EVGA e-GeForce 8800 GTX is also great news for gamers with 24-30” widescreen LCDs who have struggled with barely-playable frame rates at resolutions of 1920x1200 and 2560x1600—and have always had to compromise visual quality as a result.

XHD Benchmarks—Time to Play!

To see if all the hype about the GeForce 8800 GTX and XHD gaming is true, I dropped the EVGA e-GeForce 8800 GTX into a rig configured as follows: