Nvidia GeForce 3 details emerge
So hot it's cool
So what's so special about Nvidia's GeForce 3 introduced yesterday at - of all places, given the company's PC gaming background - MacWorld Expo Tokyo?
Digging around the Web a bit - which unearthed one or two handily leaked Nvidia slides and specs - and taking the company's official pronouncements, here's what we reckon it can do.
The new chip's key feature is its programmable rendering engine, dubbed nfiniteFX. Essentially, it gives games developers the scope to control the way the chip handles the geometry of a 3D scene and then renders it.
It's a little like Photoshop's Filters facility. With Nvidia's and other chip makers' previous technologies, games developers were forced to work with a pre-set selection of effects and operations, much as they might select one of the Adobe app's filters from the standard menu.
Photoshop has always given users the scope to create their own filter effects. The nfiniteFX engine, with its Vertex Processor (geometry) and Pixel Processor (rendering), both of which control the chip's transform and lighting engines, finally does the same for 3D software developers. Indeed, Nvidia's GeForce 3 software development kit (SDK) will come with over 100 lighting, transform and effects routines.
The Vertex Processor will handle procedural animation effects, such as facial movements, while the Pixel Processor is designed for shading effects. The latter can perform up to 36 pixel shading operations per cycle and provides "per pixel Z-connect reflective bump mapping". Golly. Together, they enable the chip to do real-time object shadows.
On the PC side, all of this will be made accessible through DirectX 8.0, which was mightily modified by Nvidia as part of the development work on Xbox. DirectX 8.0 and the GeForce 3 are essentially the mainstream versions of the APIs and accelerator technology
The GeForce 3 also adds a new anti-aliasing system, dubbed Quincunx, which samples the image at around 3.2 times the rate of the GeForce 2 Ultra. Bandwidth is maximised, we're told, by compressing data on the fly by a factor of four as it moves around the chip's innards. All of which means the chip can run Quake III at 1024x768 in 32-bit colour fully anti-aliased at 71 frames per second; at the same settings, the GeForce 2 Ultra could so 34fps. It has a theoretical fill rate of 3.2 billion texels per second.
All of this functionality is produced through 57 million transistors - by comparison the GeForce 2 has 25 million. The chip is capable, says Nvidia, of 700 billion ops per second and 76 billion floating-point ops per second. The chip's core runs at 200MHz, the 64MB (in the Mac card) of memory at 460MHz (230MHz DDR SDRAM). AGP at 2x, 4x and 8x, plus FastWrite, are all supported, as is TV output. The RAMDAC runs at 350MHz.
The core's clock speed, incidentally, is less than the GeForce 2 Ultra, which runs at 250MHz. Presumably that's to help counter the added heat-generation of the 128 per cent increase in the number of transistors. We believe the chip will be fabbed at 0.15 micron - step forward, Taiwan Semiconductor - which should help cut the heat output, too. But we'd be very surprised indeed if this part will be one of Nvidia's 'fanless wonders'.
Our sources suggest that Nvidia plans to spill the beans on the GeForce 3 real soon now, possibly as early as next week. Certainly, we can't yet confirm many of the finer points of the chip's performance and how it's achieved, but we reckon what we've got here is pretty damn close. We'll see. Whatever, we shouldn't have too long to wait for the official numbers. ®