This article is more than 1 year old
Nvidia GeForce 3 is go
Xbox marks the spot...
San Jose Nvidia launched its latest graphics chip, the GeForce 3, here in San Jose, California this evening - just hours before 3dfx's shareholders are due to meet to decide whether that company should sell off its technological and brand assets to Nvidia before dissolving.
And the company held out the prospect of further launches this year, to make 2001 one of the busiest in Nvidia's history.
As expected, GeForce 3 is an impressive step forward. Indeed, it's much further beyond the GeForce 2 than that chip was ahead of the GeForce 256, the first in the family.
That's largely down to an entirely new architecture. The GeForce 3's design centres on a programmable geometry engine - Nvidia calls it the Vertex Processor - and a programmable rendering system, dubbed the Pixel Processor. As we predicted in our GeForce 3 preview, both engines allow 3D software developers to customise the chip to the specific needs of their applications, rather than the other way round.
The upshot: developers have the scope to create a vast array of unique effects and have them run directly on the graphics chip rather than burden the host's CPU with the task. Certainly, many of the effects Nvidia describes are achievable - just - with a regular graphics chip working alongside the main processor, but with a significant performance penalty. Faster CPUs, however, will ultimately make that less of an issue.
Still, it's in the mixing and matching of the GeForce 3's many new vertex and pixel manipulation commands that will reveal how it's more than just another faster graphics chip. And that's going to be key to its success. Nvidia's demos, for instance, are impressive at a purely visual level, but fail to score in other areas.
The Max Headroom-style Persian oracle 's facial characteristics and movements may be manipulated in real time, but not enoughtto detract from the simplicity of the animation. In other words, it may look real, but it doesn't act real. Clearly, visual sophistication only takes you so far.
Not that that's Nvidia's problem - it's simply providing a broader palette than before for 3D artists to select from, not telling them how to choose. But, possibly more than the introduction of on-chip transform and lighting engines before, it's going to take some time before the new engine becomes de rigueur.
Nvidia has a couple of advantages this time round. First, GeForce 3 remains significantly faster than its predecessor, despite a lower core clock speed. That's partly due to its much-improved anti-aliasing, but mostly because of the chip's memory architecture, which is based on four controllers operating in parallel in what appears to be a version of the NUMA architecture of multi-processing fame. Essentially, each controller operates on its own bank of memory, but can access other controllers' RAM when it's underutilised and they're being driven to the max. The chip also compresses z-buffer data to maximise bandwidth, and implements z-buffer occlusion culling, essentially a technique to ensure that only visible pixels are rendered.
Second, GeForce 3, as we've said before, shares the same core design as Nvidia's Xbox graphics chip. It's also closely tied into DirectX 8 - asking which came first is a bit of a 'chicken and egg' question, it seems - so already it has a lot of support in the games development world, which should ultimately accelerate its take-up in the OEM space, which is really what Nvidia needs if its technology is to become as ubiquitous as it would like.
Company CEO Jen-Hsen Huang this evening promised that 2001 will see more product introductions than at any time in Nvidia's history. Given the arrival this year of not only the GeForce 3, but the final shipment of the GeForce 2 MX, the Xbox graphics processor, the Xbox Media Communications Processor plus a version of it aimed at the low-end PC market, you don't need to take his word for it, just do the maths. We imagine there will be one or two other GeForce 3-class chips launched before year-end too.
Don't expect any of them to feature 3dfx technology, however. Quite apart from the vote tomorrow, Nvidia won't be utilising any of it until the NV30 chip, due, according to Huang, sometime next year. ®
GeForce 3: Key Features
- 57 million transistors
- 12x12mm die size
- Fabbed at 0.15 micron
- 800 billion ops per second
- 76 billion floating-point ops per second
- Up to 36 pixel shading ops per clock cycle
- Fully anti-aliased Quake III at 1024x768 in 32-bit colour runs to over 70fps
- Quake III frame rate over 117 per cent faster than GeForce 2 Ultra