A linear increase in speed usually implies the same architecture. When Intel released the Coppermine architecture Pentium III with several different major design differences, they said that it was a new generation of Pentium III. But when Intel raised the Pentium III clock speed from 800MHz to 1GHz, they did not call the 1GHz a new generation of Pentium III because, while it was faster and there were architectural changes, it was not a new generation of CPU. We asked NVIDIA about how the T&L engine had been redesigned. Their response was this
"The GeForce2 GTS's T&L is twice as fast. In some cases the T&L performance of the GeForce2 GTS is more than double that of the original T&L engine. The T&L engine has been optimized and re-pipelined and runs at much higher clock rates."
The GeForce's T&L engine unloads T&L from the CPU and can process T&L calculations much faster than most CPUs can.
There are two main effects of the GeForce2 GTS's hardware T&L. First, it can compute T&L faster than any current consumer CPU. This allows more polygons and more detail in scenes while keeping performance at acceptable levels. Second, by having hardware calculations done off CPU, the CPU is freed up for things it does well, including running AI, sound and the game itself. NVIDIA has shown us a chart stating that, in Half-Life, graphics calculations take up 50% of the CPU's time. But with hardware T&L, graphics only take up 10% of the CPU's time, allowing the game to spend more time on other functions.
The main problem with hardware T&L currently is that not all games make use of it. A game has to use OpenGL or DirectX 7's T&L pipeline in order to use the GeForce2 GTS's hardware T&L. Otherwise, the games will use software T&L and hardware acceleration won't make any difference. This was a problem for the GeForce 256 six months ago, but nowadays, almost every game released makes use of DX7 or OpenGL, so they gain some form of T&L acceleration. Games that actually push the limits of hardware T & L capability are on the horizon.
The GeForce2 GTS uses a 128-bit memory bus linking it to 166MHz DDR SGRAM. DDR SGRAM sends data on the rise and fall of the memory clock, and therefore 166MHz DDR SGRAM acts much like 333MHz SDR SGRAM. When you do the math, you get a grand total of 5.2GBps of memory bandwidth, which is an incredibly large amount. This bandwidth, when combined with the GeForce2 GTS' powerful graphics engine, has the potential to see high frame rates at extremely high resolutions.
Consumer GeForce2 GTS cards will ship with 32MB or 64MB of memory. There is the potential for 48MB and 96MB cards as well. The architecture is capable of having up to 128MB of memory, but cards with 128MB will most likely ship only for the workstation market.