In the past year we have seen the graphics market consolidate to the point where there are only two main players at the top: ATI and Nvidia. Earlier this year, ATI responded to the GeForce 2 juggernaut with their highly successful Radeon chipset, which combined a flurry of new DirectX 8 compatible features with some of the best 2D and 3D image quality gamers have ever seen. New bandwidth saving techniques helped ATI excel at high resolution, 32-bit graphics and become a popular fixture in gaming machines. The GeForce faithful held firm, patiently waiting for Nvidia to answer back with their next generation product. The time has now arrived, and Nvidia has released the GeForce 3. Not only has it arrived on the PC, but the GeForce 3 chipset will be the core graphics engine for the upcoming Microsoft Xbox, which throws a few interesting wrinkles into the mix. How well can Nvidia meet the needs of both platforms? Will one lose out to the other, or will they both succeed in the eyes of their dedicated public? We were lucky enough to receive a GeForce 3 reference board directly from Nvidia and have run it through our standard array of tests. To find out how well it holds up and what it has to offer, feast your eyes on our latest review.

Features

We don't want to go into a super-technical analysis of the features that the GeForce 3 brings to the table, because frankly, we don't want to overwhelm and bore anyone with too much raw data. However, here are the key technologies that Nvidia has introduced in their new product and how they will benefit you.

Lightspeed Memory Architecture

These functions are designed to help minimize bandwidth requirements and are very similar to the Hyper-Z technology in the ATI Radeon. The Nvidia solution consists of four primary functions:

Optimized Memory Control
Hidden Surface Removal
Lossless Z-Buffer compression
Fast Z-Buffer clearing

The optimized memory controller allows data to be broken into smaller segments when needed. This means that instead of sending data in fixed chunks of 128 or 256 bits, it is possible to send data in smaller clusters of 32, 64 or 96 bits, for example. Less wasted data is sent over the memory bus, making the entire process faster and more efficient.

The hidden surface removal feature analyzes 3D graphics to filter out the processing of those pixels which do not actually get shown on screen because they are blocked by objects in the foreground. The computer does not spend time computing objects that it really does not have to, and as a result, less data has to be sent over the memory bus and less computing power is used in calculating the display.

Lossless Z-Buffer compression allows data to be compressed so that it takes up less space in memory. This concept is very similar to how files are compressed into ZIP archives. No data is actually lost, it is just squeezed down so that it can be transferred faster. It can then be decompressed when it arrives at its destination.

Fast Z-Buffer clearing is a highly efficient method of data-dumping that helps to get the Z-Buffer area ready to receive new data quickly. Timing is really what this is all about, since you don't want data to be flushed before it is properly rendered on screen. It is basically a dynamic method of clearing data-jams; and is somewhat similar to how traffic lights work. Some lights work on a timer, where it turns green every three minutes or so, allowing traffic to move at pre-determined intervals. However, in high traffic areas, lights may be set dynamically with sensors that can tell when traffic is particularly heavy, and adjust the delay accordingly. When little traffic is present, lanes may be flushed quickly, or when heavy traffic is present, lights may stay green a little longer. It is a complicated procedure that relies heavily on the efficiency of the driver code.

High Resolution Anti-Aliasing

This is a feature that is new to the market and a step above anything the competition has put forth to date. Like the system in the Voodoo 5500 from 3dfx, this anti-aliasing is based in hardware, but it is designed to be much more efficient. The GeForce 3 uses two different computational methods: The first is multi-sampling, which is different from the traditional super-sampling in that it is designed to eliminate redundancies, something akin to the hidden surface removal concept. The second is a proprietary algorithm called Quincunx, where pixels are blended via a cross-comparison. This algorithm is more efficient than other established methods and results in a picture very similar to 4x anti-aliasing with much less computational overhead.

Improved DVD Playback

PC DVD playback is becoming a fairly common thing, but these improvements seem targeted more towards the Xbox crowd, since every console will come with a built-in DVD player. They key items that Nvidia has included in the GeForce 3 that enhance DVD playback are:

High Definition Video Processor (full screen, full frame playback)
Hardware Motion Compensation
Sub-Pixel Alpha Blending / Composition
Hardware Scaling (up and down)

We do not include DVD playback as part of our normal video testing, but as a matter of comparison did conduct visual quality assessments using the Matrix DVD. We found video playback on the GeForce 3 to be improved over that of the GeForce 2 Pro, but lagging noticeably behind the outstanding playback offered by the ATI Radeon. During playback at 1600x1200x32, the GeForce 3 had some noticeable blur during the fight scene in the dojo and at times there were jerky hesitations. The Radeon exhibited none of these anomalies using WinDVD or PowerDVD, yet they were present on the GeForce 3 in both applications. Perhaps future versions of the PC playback software will utilize the features that the GeForce 3 has to offer to a greater degree.

nfiniteFX Engine

With most graphics cards, such as the ATI Radeon, complex effects are coded directly into the video hardware; applications can realize a dramatic increase in performance by calling these built in hardware functions as opposed to writing them in software. The downside is that applications are usually limited to only those functions that are built directly into the hardware. Nvidia has taken a somewhat different approach with the GeForce 3. The nfiniteFX engine is designed to be fully programmable, with written functions receiving hardware assistance where possible. While somewhat slower than dedicated hardware functions, it is still substantially faster to have programmable effects that are hardware assisted instead of solely processed in software.

The two key areas of programmability are the Vertex processor and the Pixel processor. These systems can help to dramatically increase performance for complex vertex and pixel manipulation without limiting programmers to a fixed set of functions. This programmability may be key to attracting developers from different platforms with different gaming engines. By allowing each developer to custom-configure their environment, it will allow them to build upon existing projects while promoting the possibility of more advanced and more complex environments. Nvidia is providing development kits to help speed acceptance, but it will still take some time for developers to take advantage of this new flexibility. However, since the GeForce 3 is at the heart of the upcoming Xbox gaming console, it is very possible that PC gamers may see a flood of innovative titles coming their way sooner than they might otherwise expect. Some games may be ported directly to the PC from the Xbox, while others may be developed using the engines created for the Xbox console. Either way, it could be a boon for gamers and developers alike.

The bulk of the features mentioned above help bring Nvidia neck and neck with their main competitor -- the ATI Radeon -- in terms of features and functionality. The exciting nfiniteFX Engine and their improved anti-aliasing technology pushes the envelope even further, and now establishes them as a leader in more than just raw power. Nvidia has sought to round out their development, combining finesse with brute force. How successful they will be depends on how quickly these new features are adopted and how easy they are to implement. But with Nvidia's track record, it will be hard to bet against them.