The GeForce 200 GTX
A new generation, a new name. Using the “10,000” place for a series name seems to bother GPU manufacturers. Whereas ATI solved the problem by introducing Roman numerals, Nvidia has completely changed its card nomenclature. Now we have the GeForce 200 GTX. But there’s a question we just have to ask: What happened to the GeForce 100 GTX?
As we’ll see, Nvidia is introducing a new architecture, but they’re not exactly starting afresh. While the genesis of the G80 had started with a blank page, the architecture has proved to be extremely efficient. The goal for the GT200, then, was to correct all the architecture’s minor youthful errors and better equip the circuitry for the games of the future. Like what the G70 was to the NV40, what we have is a raft of small improvements, but above all a big leap in processing power.So, where the 8800 GTX offered a floating-point processing power of 518 GFlops, the GTX 280 has close to a Teraflop – an impressive 933 GFlops. And in reality, the spread is even wider, since that figure of 518 GFlops for the G80 is based on the execution of three floating-point operations per cycle (one MAD and one MUL) – which, due to the G80’s limitations, was impossible to achieve in practice. With the GT200, Nvidia guarantees, with tests to back them up, that those problems have been solved. To almost double the processing performance of its previous GPU, Nvidia substantially increased the number of multiprocessors – from 16 to 30.
Nvidia and AMD agree that the games of the future will need significantly higher arithmetical power compared to texturing power, and consequently it’s no surprise to see that the number of texture units shows a more modest increase. From 64 on the 9800 GTX, the GTX 280 has moved to 80 (and if you compare the 8800 GTX to the GTX 280 this time, we go from a theoretical arithmetic instructions / number of texels filtered ratio of 14.1:1 to 19.4:1). Let’s look at what that means right away.