RSS Feed

HCW Tech Blog

For the latest info on computer hardware, tech, news, video games, software tips, and Linux, check out our new improved front page: HCW Tech Blog

Reviewed by: Bryan Pizzuti [04.22.02]
Manufactured by: Chaintech


Features - The GeForce3 Core

The GeForce3 GPU forms the basis of GeForce3, GeForce3TI200 and TI500 cards. It brought a LOT of advances to the Geforce family, which I will summarize here. Anyone looking for more detailed information can find it at

Shadow Buffer

Technically, this is part of the Transform and Lighting engine (specifically the Lighting part). Basically, what this does is process all of the surface and lighting data for a scene and automatically generate realistic shadows based on this data. It then stores the shadows in a buffer, where they can be accessed as ordinary textures and placed into a scene. Until recently, shadows had to be manually processed, and weren't always accurate. For instance, if you've ever played a flight sim, you may have seen a shadow directly under your plane. But was it dependent on where the sun was? And, even more importantly, did your plane's fuselage cast a shadow on the wings? Because shadow processing can be very complex, developers tended to keep it to basics in there games, however, NVIDIA's Shadow Buffer can give developers a great shortcut to richer, more realistically rendered scenes.

Lightspeed Memory Architecture

One of the greatest problems with video cards has NOT been the speed of the processor chip, but what's become known as the "memory bottleneck." As more and more polygons and textures have found their way into games, graphics processors have risen to the challenge. But memory faced an even greater challenge, because the amount of data that these GPUs could process was, and still is, enormous. And memory technology has been unable to make the incredible leaps to keep up. To remedy this, the "Big 3" video card manufacturers have each come up with their own way of trying to reduce the load on the graphics card's memory bus. PowerVR's method has gained recognition as being one of the most innovative, by basically taking everything that won't ever be seen by the user, and throwing it away. A description of their methods can be found in our 3D Prophet 4500 review, with special guest star Kevin Bacon (Thanks Kev!). ATI's method involves a lossless, high-speed compression for all of their data called Hyper-Z, and basically works like ZIP or RAR built into the card. More about ATI's HyperZ can be found in our Radeon 7500 review.

NVIDIA actually implemented something similar to PowerVR, by throwing away some of the data that won't be seen, and therefore doesn't need to be processed, in their Z-Culling. They also followed in ATI's footsteps and implemented their own lossless Z-buffer compression. But they also had to do something with the memory bus, and instead of trying to whip it into going faster, they decided to make it more efficient instead. The result is their Lightspeed Memory Architecture, also known as the Crossbar Controller.

As you can see from the diagram above, the Crossbar controller doesn't use a single 128-bit wide memory path. Instead it uses 2 memory paths that are 64 bits wide and are interleaved between the individual RAM chips (the controller is actually 2 independent memory controllers). And yes, this works, because data is sent back and forth from the memory in "chunks" that are the same size as the width of the bus, and double that with a DDR bus. So to send 50 bits to the GPU would actually take up 256 bits of bandwidth with an ordinary 128 bit DDR memory bus. But with NVIDIA's Crossbar technology, it only takes up 128 bits out of the available 256; effectively only needing half the bandwidth that it used to need. Now, that might not seem like a lot of difference numerically, but now try multiplying that savings by about a million transfers PER SECOND, and it adds up very quickly.

The nFinite FX Engine

This is NVIDIA's cool new name for their processing engine, and it's got several major features of it's own, in addition to enhancing hardware T&L operations.

3D Textures

When textures have been applied to polygons in the past, it always had the effect of attaching a painting to a ball or cylinder; meaning they were just "painted on" 2-dimensional objects. To give them any sort of 3D effect, such as facial features, or the pits of a golf ball, developers had to use optical illusion; basically painting the dimples onto the golf ball. So as far as the games and GPUs were concerned, you could see dimples, but when you put your hand on it, it was a flat surface. But NVIDIA's 3D texture technology adds a third dimension to the texture "painting" allowing textures to have depth as well as width and height. Not only does this make textures more realistic, but it also reduces the load of having to process all sorts of "optical illusion" effects to make textures appear to have depth.

Vertex Shaders and Pixel Shaders

Previously only server farms could do these massively intensive calculations to perform operations on the vertex of a triangle or apply shades to each individual pixel on a surface. But now, the nFiniteFX engine allows developers to make use of these functions, most of which no user has heard of. Nor do you need to know the nitty-gritty details of their functions, but you should know in general. Basically, these technologies combine to bring an incredible increase in the detail and realism levels of polygons and their surfaces, by applying high-end calculations to the vertexes of the triangles, and their associated textures as well. For example, this can make it easier to construct a realistic face, complete with dimples, scars, pimples, pockmarks, and whatever else, give these features realistic shading, and then perform operations on the polygons in order to pull the face into a smile, or a frown, or flare the nostrils. Previously, this was VERY calculation intensive, but the nFiniteFX engine provides a shortcut, allowing these effects to be used MUCH more often in a scene. Instead 0of the focal character in a room of 20 showing expression at specific times, imagine all 20 people in the scene with their lips flapping, eyes blinking, smiling, and whatever else. It definitely has the potential to increase immersion factors.

Next Page: (3)