NVIDIA GTX480 and 480 SLI review
Author: Luka Rakamaric
Date: 26 Mar 2010
After six months of losing to ATI in all benchmarks, NVIDIA is back with a vengeance. The long awaited GF100 GPU is finally here, in the form of NVIDIA's GeForce GTX 480 card. It is built on a new NVIDIA architecture called Fermi, which should not only enable great gaming performance, but should also provide the ability to use its huge processing power in various other scenarios.
You can find more on the Fermi architecture in our separate article, while here we will focus more on the graphics performance of the new GPU.
The GF100 (G for Graphics, F for Fermi and 100 to denote that it's the high end GPU of the GF family) is a 3 billion transistor 'monster', built in a 40 nm production process. 3 billion transistors is more than twice what we had in the GT200 and 50% more than what ATI has in its current generation with Cypress (5870) . 3 billion will get you to 512 stream (shader) processors, but the GTX 480 will only have 480 of them active. This was done to improve yields, as with 3 billion transistors there's a big chance some of them will be faulty. With this 6.25% reduction in required operational shaders, the yields are automatically increased by a nice factor.
The 512 cores are divided into 16 sections called Streaming Multiprocessors, and each of those contains 32 CUDA processors. That is four times more than what we had in previous designs. By using a scalar architecture, NVIDIA ensures that all of their cores can be used simultaneously with maximum load, unlike ATI, which does have more cores, but with a superscalar design in groups of five.
Let's see some pictures, stripped down and "naked":