NVIDIA has officially launched its latest GPU, the GTX 680, which is based on an all-new graphics chip architecture codenamed Kepler.
Built on a 28nm process, the GTX 680 has 1,536 CUDA cores (CUDA is nVidia's parallel computing architecture) and runs at a base clock of 1,006MHz.
It also has 2GB of 256-bit GDDR5 memory and works on the PCI Express 3.0 interface.
The GTX 680 has improved PhysX support and next-generation anti-aliasing features that result in more complex high-density DirectX 11 tessellations and smoother edges, the company said.
While nVidia claims up to 40% performance increase over its nearest competitor, the company has managed to produce a card which draws 30% less power and is 52% quieter.
This is achieved not only from the 28nm process, but also thanks to a new design which reduces the amount of control logic (the part which handles the data traffic) in each cluster of stream processors (which handle the actual computation).
Reducing the amount of control logic also greatly reduces power consumption.
The GTX 680 also has a feature similar to Turbo Boost in Intel CPUs, i.e. the GPU is able to dynamically increase its clock speed for better performance depending on need and power consumption at any given time.
In fact, the GTX 680 is so power efficient that it runs at a TDP (thermal design power) of only 195W, and the reference board design only requires two six-pin PCIe power plugs instead of the eight-pin and six-pin combination of most modern high-end graphics cards.
The GTX 680 is available now from various nVidia hardware partners, such as Asus, Palit, and Gigabyte.