The PC graphics industry is bitterly competitive. That's a fact. Here's another: to have any chance of succeeding in it you have to be first or you have to be fastest.

So here's the rub for ATI : its new Radeon HD 2900 XT flagship video card, officially launched today, is neither.

Normally we'd string out the final analysis until the last paragraph. But the Radeon HD 2900 XT is not a normal product launch. For starters, it's well over six months late. That alone is a court martial offence in the context of the warp speed graphics market.

However, it's also unusual for being launched with a vaguely affordable price tag. In recent years, the launch prices of new flagship graphics chips have been creeping inexorably upwards. Anything under £400 would therefore look cheap, but the HD 2900 XT is rolling out for around £250.

Precarious positioning

That's really significant because it means ATI has intentionally positioned this new graphics board to take on the second rung products produced by its main rival, Nvidia . It has effectively been forced to make the humiliating decision not to compete at the highest level.

To understand why, you'll have to get down and dirty with some of the finer points of the Radeon HD 2900 XT's architecture.

Like Nvidia's competing DirectX 10 class graphics chips , the 2900 XT sports a fully unified shader architecture. In simple terms, the chip is powered by a highly programmable floating point array capable of seamlessly switching between pixel, vertex and geometry processing.

The basic idea is to maximise the efficiency of the chip and minimise the chances of any functional units remaining idle during graphics processing.

It's the same concept used by Nvidia for the Geforce 8800 series of GPUs. The 2900 XT also mirrors its Nvidia competition in terms of manufacturing process. Both are based on 80nm technology.

Gigantic GPU

Likewise, it's another massive chip. If anything proves that ATI had originally intended this to be the fastest graphics processor on the planet, it's the gigantic 720 million transistor count. That's some 40 million more than the GeForce 8800.

The 215W power consumption also indicates this is a graphics card that really ought to be duking it out for the title of world's fastest graphics card.

Nevertheless, it's also packed with some extremely impressive new technology. The world's first 512-bit memory bus and a similarly record breaking core clockspeed of 740MHz are just a few items from the monumental spec list.