Nvidia GeForce GTX 680 Launch Review - PAGE 1

Today marks an important step in the evolution of the graphics card market, as Nvidia is releasing their first new architecture since 2010. The new architecture is dubbed “Kepler” and utilizes a 28nm node, sports a redesigned SMX architecture and is built on three core principles of making a faster, smoother and richer gaming environment.

Starting with the core, Nvidia has built on the foundation of the Fermi architecture by utilizing a similar CUDA core design, but Kepler uses an entirely new geometry pipeline to improve tessellation as well as improved displacement mapping and compute operations to better performance across the board. Additionally, the Kepler architecture eliminates the “hot-clock” design used by Fermi, so the CUDA core will no longer run at double the GPU clock speed. This allowed Nvidia to instead triple the CUDA cores inside the core, without the power consumption spiking through the roof. However, the tradeoff is the CUDA cores inside Kepler are not as fast as the ones seen in Fermi, so we can’t make a direct comparison between the two. Some Nvidia reps have suggested that 512 Fermi cores equate to around 768 Kepler cores, give or take a few.

The GPU also features an entirely new clock design that utilizes both a stock and boost clock speed. The stock speed is considered a guarantee of the lowest GPU frequency used while gaming, while GPU Boost dynamically increases the clock speed above the stock frequency to ensure the best results. GPU Boost works at both a hardware and software level, and monitors the GPU's power consumption and heat level before dynamically boosting the clock speed. While the GPU boost frequency of the GTX 680 is set at 1058MHz, it is possible to see clocks beyond 1100MHz as long as the conditions are acceptable. GPU Boost is also considered the average speed the GPU should run at under most load conditions.

Nvidia has also entirely reworked the memory subsystem of Kepler based boards. This allowed them to push the memory frequency up to 1500MHz (6000MHz effective). At this speed the memory is able to achieve a bandwidth rating of 192GB/s, even while running on a 256-bit interface. In our opinion this an important piece of the Kepler design, as it allows Nvidia to create the board at a lower cost than if they were using a larger fame buffer and memory interface. In turn, the design allows them to price the GTX 680 lower than the HD 7970. If Nvidia’s claim of the GTX 680 being 10% faster than the HD 7970 holds true, AMD is going to have to drop their prices if they want to stay competitive.

On top of all the aforementioned specifications, Nvidia builds upon this already impressive feature set with support for 3D Surround via single graphics card, improved anti-aliasing features and improvements to PhysX. Obviously we have a lot to cover, so let’s move on and see what makes this beast tick, and more importantly, how it performs against AMD's Radeon HD 7970.