The card T&L par excellence: GeForce 256
It was already an hour... Well, convinced we are, now: what cards that they use T&L can we acquire? Since apart from the GeForce, which exist of numerous manufacturers (Creative, ASUS, Leadtek, ELSA...), little more.
We have the cradles in the chip Savage 2000 of S3, that although it is not not great less at a height of the GeForce, also it is true that it is bandaged by a small fraction of the price of this one; and it seems that ATI is ready to use this technology in a future, starting by a chipset with integrated graphic checker. The only one who does not think to do cards with T&L for the time being is 3Dfx, that also it spends for bad moments due to the relative antiquity of his Voodoo 3...
But good, since the Savage 2000 scarcely is on the market, and that T&L is a nVIDIA technology, we are going to center on the specifications of the GeForce:
Processor (GPU) with 23 million transistors:
The whole monster of chip, which also lows the price of the costs on having integrated in the same processor the technology that allows to apply T&L and the render units.
Support for T&L:
Major graphic potency with major detail.
Independent Pipelined QuadEngine ™:
The calculations must be done by the following order: transformations, lighting, triangle setup and renderizado; nevertheless, when we have already transformed certain points, we can be applying the lights, be preparing the triangles and even renderizar some píxels.
The chip GeForce has 4 units of render of 64 independent bits, what allows him to segment (pipeline) the work. It accelerates the speed because instead of doing the things sequentially, there follows a species of parallelism (the modern microprocessors work in the same way).
These units are capable of renderizar up to 4 píxels of 32 bits for cycle of clock (the double that the TNT2). These 64 bits are distributed in: 24 bits for the color, 8 for the channel Alpha (transparences), 24 for the Z Buffer and 8 for Stencil Buffer (that serves to create effects as the shades of Quake III).
AGP 4x with Fast Writes:
The last type of access to memory and theoretically the best as for speed refers, something improved for nVIDIA. Nevertheless, the AGP 4x only is in the basic badges with chipset i820 or Apollo Pro 133rd, and all the tests seem to indicate that in practice an AGP can still be enough perfectly 2x.
RAMDAC to 350 MHz:
He supports refreshments of screen of 75Hz to 2048*1536 (a little that leaves of the possibilities of almost any monitor of the market, even if it is 21 inches).
Fill rate of 480 million píxels:
With leaked bilinear and multitexture, what implies approximately 3.8 GigaTexels in second (like much).
15 million triangles / second:
That (as maximum) he can handle, transform, illuminate, prepare and renderizar.
Now a note on the functioning when multitexture is done (to apply 2 textures or more to the same pixel). With the GeForce, we have 2 options: or renderizar the same texture in 4 píxels different or 2 different textures each one in 2 píxels (two textures for pixel) in every clock cycle; why cannot we apply 4 different textures in 1 pixel how in the Savage 2000? Probably because for the design of the new card it did not break of a completely new design, but based on that of the TNT2 (almost how to take two chips TNT2 and to join one one next to other, a species of architecture "TwinTwiNTexel").
Anyhow, at the moment, this is not too important, since the current games as much apply 2 textures to the same pixel (or I object), nevertheless this can change in a future (earlier the normal thing was to apply only one texture, but from the Quake II...). Clear that for when it should change, the GeForce will be considered to be an almost prehistoric card, it is not necessary to worry.
32 Megabytes of Memory:
Sufficient for most of the mortal ones, also he supports SDRAM, SGRAM and DDR-SDRAM. With different prices and yield, of course; the best is the DDR-SDRAM.