NVIDIA Announces Fermi GeForce GTX 480

NVIDIA announced the GeForce GTX 480 in March 2010 as the first flagship consumer card based on its Fermi architecture. Fermi was designed not only for graphics but also for increasingly important general-purpose GPU computing.

The architecture expanded CUDA capabilities with stronger double-precision support, a more conventional cache hierarchy, and improved scheduling for compute workloads. The GF100 chip was extremely large and power hungry, and the GTX 480 became notorious for heat and noise.

Despite those problems, Fermi was important for CUDA. NVIDIA was turning the GPU into a parallel-computing platform used in simulation, scientific computing, and later machine learning. The GTX 480 showed both sides of that strategy: enormous computational capability and the engineering difficulty of delivering it within consumer power limits.