NVIDIA Launches the Kepler GPU Architecture

NVIDIA launched its Kepler GPU architecture in 2012 with products including the GeForce GTX 680. Kepler was designed to improve performance per watt compared with the preceding Fermi generation.

The architecture reorganized CUDA execution around larger SMX units and relied on more CUDA cores running at lower clock rates rather than very fast shader clocks. NVIDIA also introduced GPU Boost, which adjusted frequency dynamically according to power and thermal conditions.

For CUDA users, Kepler added features such as Dynamic Parallelism in later compute-focused chips, allowing GPU kernels to launch additional kernels without returning control to the CPU. Kepler’s efficiency mattered beyond gaming because GPU computing was expanding into scientific workloads, simulation, and early deep-learning research.