nFiniteFX Engine

The NVIDIA Infinite Effects GPU is a breakthrough in GPU technology. It features the nFiniteFX engine, which provides the first ever fully programmable graphics engine. This opens up a whole new realm to developers by allowing them to design their own effects for games instead of being restricted to a certain pre-defined set. The nFiniteFX engine actually encompasses several key components, which I am going to elaborate on.

Programmable Vertex Shader

First off I want to refresh everyone's memory as to what a vertex is. As you may or may not now, graphics are basically made up of a bunch of triangles connected together. Quoted straight out of the dictionary, a vertex is "The point at which the sides of an angle intersect." With that in mind a vertex shader controls the position and look of objects (triangles) defined in games. Having a programmable vertex shader gives designers the ability to write their own set of instructions to dynamically change objects in games. Best of all since the GPU is specifically engineered for graphical equation computation; these results can be achieved with little or no frame rate reduction.

Programmable Pixel Shader

A pixel, as I am sure most of you know, is one of the many small dots outputted to your screen from a graphics card. When combined together all the millions of pixels make an image. With that said, a pixel shader provides developers with complete control over light, shade, and color of each individual pixel. Its programmability allows developers to create custom instructions to effect shading or lighting on a pixel or group of pixels in a game. The pixel shader on the GF3 can modify up to 36 pixels on each pass whereas the GF2 was limited to 7.

LightSpeed Memory Architecture

With prior graphics card, bandwidth was the biggest drawback preventing cards from rendering high-quality textured images. Even with the introduction of AGP 4x, only a total of 1GB/sec of bandwidth is available to the graphics card. While this may seem a lot, rendering high-quality images easily consumes this bandwidth. NVIDIA has released several technologies to help reduce this bandwidth problem: High Order Surfaces, Crossbar Memory Controller, Lossless Z Compression, and Visibility SubSystem: Z-Occlusion Culling.

High Order Surfaces - With traditional graphics cards, in order to obtain a curved surface, many triangles were used to get the fine curved edges. With High Order Surfaces however, simply defining "control points" can create a curved surface. This helps reduce bandwidth because with traditional graphics, hundreds of points were required to be passed to the GPU to create a curved surface. By use of control points, a very small and efficient formula can be passed to the GPU to easily render the same graphics.

Crossbar Memory Controller - With traditional memory controllers that utilize DDR memory on the board, a 128bit controller would be used to access data from the DDR memory. When accessing the memory, the controller would take 256bit chunks of data. While this would be good in some scenarios, there is a lot of times where there isn't 256bits of data available to fill the space. This leaves a lot of wasted space that could be used. The Geforce3 corrects this problem by utilizing four separate memory controllers which access the data in 64bit chunks, better utilizing the frame buffer bandwidth. Each controller has full access to all memory-providing load balancing for the system.

Lossless Z Compression - With the use of the new Lossless Z compression, Z data is compressed in a 4:1 ratio allowing for better utilization of the Z-buffer. The compression and decompression of the Z data takes place in real time and causes no graphic degradation.

Visibility Subsystem: Z-Occlusion Culling- In traditional graphics systems every pixel of a game is rendered on your screen even though you physically can't view it do to the fact of something in front of it. For example, a character that is standing in front of a wall, you cannot see the portion of the wall that the character is covering yet it is still rendered. With Z-Occlusion the GPU uses a technology, which attempts to determine if a pixel is viewable by the user. If it determines that it is not, it does not display it. This reduces overall bandwidth greatly.

High Resolution Antialiasing
Antialiasing is the concept of reducing what are commonly known as "jaggies" in graphics. For example if antialiasing is disabled the edges of objects will appear jagged instead of being smooth as they should be. In previous graphic systems, high-resolution antialiasing caused a high reduction in frame rates making most games unplayable at higher sampling rates. With NVIDIA new QuinCunx technology however, high resolution antialiasing is now available at a fraction of the frame rate cost as with previous graphics cards. Quincunx technology offers sampling comparable to that of 4x modes on previous graphic cards with only a 2x hit in performance.

The Visiontek GF3 comes in a very nice looking box and includes a CD with the drivers on it. The Visiontek GF3 also features TV-out.

Testing
For testing out the GF3 I used several games and applications: 3DMark 2001, Aquanox, Quake3 Arena, and Vulpine GLMark.

For a comparison, I used the Geforce2 Ultra.

Other Notes:
Quincunx with the Geforce3 is equal to that of 4x AA on the Geforce2.

System Specs:

1Ghz PIII
128MB PC133 SDRAM
Maxtor 80GB 1394 External Storage
Maxtor 60GB 7200RPM HD
Belkin 1394 PCI Adatper
Adaptec Fireconnect 4300
Netgear 10X PCI Adapter Card
3COM 10/100 PCI Network Card
DirectX 8.0a
NVIDIA 12.0 Drivers for both the Geforce2 Ultra and Geforce3

First off I want to show two closeup shots, one with the Geforce2 at 4x antialiasing and the other with the Geforce3 with Quincunx enabled. From the screenshots, you should be able to see that the QuinCunx actually does a better job at antialiasing then 4x with the Geforce2

Quake3 Arena

Notes:

High - Uses the built in High Quality pre-defined setting and only modifies the resolution.

Custom - Maxes out all the graphics settings and uses vertex lighting

(Demo2 High)

(Demo1 Custom)

(Demo2 Custom)

Vulpine GLMark

GLMark Note: 1600x1200x16 did not render properly on the Geforce2 Ultra

AquaNox

Note: Pixel shading was disabled for the Geforce2 since it does not support it.