Nvidia Corp. on Thursday unveiled the first member of its code-named Kepler family of graphics processing units, the GeForce GTX 680 graphics card. The new solution is aimed at high-end gamers and is based on the GK104 graphics processor, which is not the top-of-the-range Kepler chip that will be available this year, but which delivers performance that in many cases outshines that of competing AMD Radeon HD 7970.
"The Kepler architecture stands as Nvidia's greatest technical achievement to date. It brings enormous performance and exceptional efficiency. Gamers will love the GTX 680's performance, as well as the fact that it doesn't require loud fans or exotic power supplies," said Brian Kelleher, senior vice president of GPU engineering at Nvidia.
When designing Kepler architecture, Nvidia attempted to keep structure of the graphics processor similar to that of Fermi. As a result, the GK104 processor resembles Fermi in many ways, but there are a lot of internal changes that were made to boost performance and power-efficiency. For example, Nvidia to completely redesigned main building blocks of Kepler/GK104, including graphics processing clusters (GPCs), stream multiprocessors (SMs, which are now called SMXs), geometry processing units and so on. The chip now consists of 4 GPCs, 8 SMX, 1536 stream processors and so on.
Among other things, the developer decided to no longer clock compute cores at double frequency, but to make them work at the same clock-speed as the whole chip, at the same time increasing their number per SM to 192 from 32 (in GF110) or 48 (in GF114) in case of Fermi architecture.
Nvidia Kepler-based GK104 processor is generally a successor of the code-named GF114 chip (which can be called Fermi Light, as it does not support certain features and architectural peculiarities, such as high-speed double-precision GPU compute), not the code-named GF110 (which is fully-fledged Fermi with support for professional and compute capabilities). Thanks to 28nm process technology and considerably improved efficiency, the chip with 3.54 billion transistors inside is only 294mm2 large, which is below not only the GeForce GTX 580 (GF110, 520mm2), but even the Radeon HD 7970 (Tahiti XT, 365mm2) and GeForce GTX 560 Ti (GF114, 360mm2).
The GeForce GTX 680 powered by GK104 features 1536 stream processors, 128 texture units, 32 raster operating units as well as 256-bit memory bus. The chip is clocked at 1006MHz, whereas 2GB of memory operate at whopping 6.008GHz. The novelty fully supports modern high-end features like DirectX 11.1, OpenGL 4.2, OpenCL 1.2, stereoscopic-3D, 4 multi-monitor capability, PhysX, 4-way multi-GPU, PCI Express 3.0 and so on.
The GeForce GTX 680 GPU is available now from the add-in card suppliers, including Asustek Computer, Colorful, EVGA, Gainward, Galaxy, Gigabyte, Innovision 3D, MSI, Palit, Point of View, PNY and Zotac. Expected pricing is $499.