3D Chronicles, Part I: 1995-1997

The year of 2007... The long-awaited Microsoft Windows Vista arrives almost six years after Windows XP. Its radically overhauled model of interaction with computer devices makes it possible to implement a brand new approach to 3D graphics with DirectX 10 API. The main element of this approach is a step towards the unified architecture of modern graphics solutions: the entire line of discrete graphics cards from AMD and NVIDIA has been reoriented to the new products supporting DirectX 10. The first integrated solutions are to arrive in the nearest future. Out of doubt, this contributes to more flexible usage of GPU resources, not only to graphics processing. But our modern 3D corporations have gone a long evolutional way to enter this level.

This series of articles will offer you a tour tracking back the history of 3D graphics companies.

Before we proceed, let's think about the need for devices dedicated to 3D graphics processing. Until the 90s, such computing was up to huge computer systems - some animation frames took days to render. Even if the hardest compromises were made as far as graphics quality and speed were concerned, playing such graphics on a single computer was out of the question. Some of the reasons were the lack of proper software environment for efficient CPU usage as well as peculiarities of its operations: games were required to compute sounds, physics, etc, and processors were designed for universal code, not 3D graphics. The latter rendered CPUs inefficient for computing graphics, the first generation of primitive graphics devices (we'll speak about them below) easily outperforming it manifold.

But even such promising devices had a lot of problems and went flop. Market participants were destined to change.

Approaches to 3D graphics: so good and so different

It looks obvious to us today that there are geometric primitives consisting of polygons (tetrahedrons), textures applied to them (it resembles the process of pitching a tent :), and there are shaders to manage the above elements... But these problems were really pressing at the time, and the answers were not as clear as required.

Diamond Edge3D based on NV1, or you must spoil before you spin

NVIDIA NV1 had actually become the first 3D accelerator for the mass market. The Diamond Edge3D (based on NV1) launched in 1995 was a sterling multimedia system: its PCB contained sets of logical elements responsible for 2D, 3D, video, and audio. The 3D component consisted of 200 thousand transistors (FYI: the NVIDIA GeForce 8800 GTX contains almost 700 millions of them), which were intended for quadratic patch geometry models. It was an ambiguous move, considering that the industry had chosen polygonal graphics. Besides, 2D graphics quality couldn't compete with other options at that time, especially in DOS. And its audio quality was mediocre at best. Being quite expensive, it failed to win the market and remained just a museum specimen - the first experiment not only in 3D, but also in creating a generic multimedia device.

Let's return to the old times, when this new class of devices had been just appearing. Year 1996 had become much more progressive for 3D graphics after the NV1's fiasco. Three companies offered their products in this unknown field - PowerVR, Rendition, and 3Dfx. Each of the new products brought its own vision of polygonal 3D graphics. For example, PowerVR stored a geometric image of an object and split the future image into a grid consisting of rectangular tiles. Then the card checked which part of the geometry image would be visible to a user in each tile. After that only visible tiles were processed. It helped save computing resources by not wasting them on invisible surfaces. On the other hand, it resulted in much more complex coding procedures, so it was increasingly difficult to support the product. Later on the company discontinued it support for the first products and its own API.

Overdue Verite 2200 from Rendition

What concerns Rendition, this company suggested using a 3D accelerator for some operations (transforming and lighting) instead of a CPU. That was why some transistors of the Verite 1000 were used for an additional unit. Its pixel rate was 25 million pixels per second, twice as low as in its main competitor - 3Dfx Voodoo Graphics. It did have a negative effect on its performance. What killed the Verite 1000 was its strange PCI operation: on one hand, the unique data encoding system of the Rendition raised data transfer efficiency to a new level; on the other hand, it limited its compatibility with motherboards (which had to support this function.) The next generation, Rendition Verite 2x00, did not have these problems. However, it was overdue, as the market had been already conquered by the 3Dfx, and its software problems remained. Rendition was not ready for Microsoft DirectX. Besides, its own API Speedy3D had only limited support.

Development and support for proprietary APIs was relevant at dawn of 3D graphics. You can see it, if you examine universal APIs of that time - Microsoft DirectX and SGI OpenGL. It was an embryonic stage per se, when available APIs couldn't satisfy ambitions of chipmakers. Each of them strived after its own common denominator for the entire 3D industry. This desire to create complex individual products was the reason why the plain 3Dfx Voodoo Graphics was the winner.

The first logo of 3Dfx Interactive: you could see this very logo on boxes with the latest 3D games