GeForce 3 - 7 way shootout (1/42) |
Even after the chip was finally officially announced, test samples were hard to come by and also took quite a while to make their way to the usual sites. The few early sample boards that did pop up here and there were declared "non-final" and "unstable" by NVIDIA and their performance openly questioned, probably in an attempt at damage control should these renegade cards not perform as expected (especially after the hype they had been receiving). It took until mid May for GeForce 3 boards to finally become available in reasonable quantities. Meanwhile, NVIDIA hasn’t offered any explanations for this delay, and by the looks of things we shouldn’t be expecting one anytime soon either. The most likely reasons are immature/unstable drivers and/or problems with the early board designs. To make things even more confusing, there are currently two different revisions of the GF3 out and about on boards. The earlier A3 revision still bears the original "NV20" on the chip, while the newer A5 carries the GeForce3 nametag and can be found on slightly modified boards. Two of the cards tested here still use the A3 stepping, although this does not seem to have any impact on performance or stability compared to the A5. (Find out how to tell the two apart here.) The Technology There have been literally hundreds of articles about the technology behind this chip, one of the most interesting being this one. Nonetheless, lets take a quick peek at the most important features of NVIDIA’s newest wonderchip: Weighing in at 57 million transistors, this chip has more than doubled in complexity compared to its direct predecessor, the GeForce2 (27 million). The processing power has also increased from 46 to 76 GigaFlops – while remaining at identical clockspeed. Nonetheless the chip is barely larger than the GeForce 2, thanks in large part to its smaller 0.15µ manufacturing process.
The GeForce 3 isn’t a simple die-shrink of the GeForce 2, however - it is a completely new design. The new buzzword is "programmability", embodied in the chip’s new pixel and vertex shader, or "nFinite FX Engine" (as NVIDIA’s marketing people have dubbed it). While the GeForce 2’s T&L engine could render polygons (triangles) within a plane or space, it wasn’t capable of performing polygon calculations. The new Vertex shader is fully programmable, though, giving it the ability to squash, stretch or otherwise deform objects – all without any help from the CPU. Basically, whenever it is used, the vertex shader completely replaces the chip’s T&L engine, which is nonetheless still present. In effect, this makes it an either or feature: Either the T&L engine or the vertex shader.
|
Copyright: 12.06.2001 - RIVA Station 2001 - Lars Weinand - Translation by Benjamin Kraft! URL of this Article: www.rivastation.com/7gf3cards_e.htm - If you want to link to it, please use this URL! :-) |
Originally slated for release late
last year, the GeForce 3 was mysteriously delayed until the end of February, with NVIDIA
refusing to comment on the delay. Naturally, this got the rumor mill churning, spawning
reports ranging from design flaws in the chip to NVIDIA using the delay for more
strategically motivated reasons.