The new GeForce GTX 680 is the first GPU based on “Kepler” architecture. NVIDIA’s Kepler architecture builds on the technology first established in 2010 with the launch of the GeForce GTX 480, the first GPU using the “Fermi” architecture in its nomenclature. The big thing about Fermi is that it featured an entirely new parallel geometry pipeline optimized for tessellation and displacement mapping, as well as a new compute architecture that offers faster context switching and improved performance for atomic operations. The new Kepler GPUs are composed of many different hardware blocks that make up the different parts of the GPU. Each performs specific tasks that dynamically improve the way your GPU performs and functions. Some things however remain the same such as the GPC continues to be the dominant high-level hardware block in Kepler. THE GPC uses its own dedicated resources for rasterization, shading, texturing, and compute. This means most of the GPU’s core graphics functions are performed inside the GPC.
Inside the new Kepler GPC resides the next generation Streaming Multiprocessor (SMX). SMX not only provides more performance than Fermi’s SM, but does so while consuming significantly lower power. Most of the key hardware units for graphics processing reside in the SMX, which now supports 192 cores. The SMX’s CUDA Cores perform pixel/vertex/geometry shading and physics/compute calculations. Texture units perform texture filtering and load/store units fetch and save data to memory. Special Function Units (SFUs) handle transcendental and graphics interpolation instructions. Finally, the PolyMorph Engine handles vertex fetch, tessellation, viewport transform, attribute setup, and stream output. In addition to SMX, Kepler also boasts an improved memory interface. NVIDIA engineers completely reworked the memory subsystem for Kepler, enabling much higher memory clock speeds compared to prior GeForce GPUs. As a result of this effort, GeForce GTX 680 achieves memory speeds up to 6Gbps, the highest speed of any GPU in the industry.
SMX features substantially more shader, texture, and geometry processing power compared to the previous generation Fermi, while the double speed processor clock present in prior Ge Force GPUs has been eliminated in favor of more CUDA cores operating at the lower-speed graphics clock. Running the CUDA Cores at 2x clock allowed us to extract the most performance in the least amount of die space for Fermi, but the downside was higher power consumption. With Kepler, each SMX processor contains up to 192 CUDA Cores; that’s 6x the number of cores per SM compared to Fermi, providing 2x the perf/watt equation and allowing much better performance per clock cycle. As far as power goes the new GTX 680 takes a step in the right direction by eliminating the need for a Six and Eight Pin Connector required by AMD's top dog the HD7970 by allowing Two Six Pin connections and less TDP by 25W as the AMD HD7970 requires 220W and the new GTX 680 only requires 195W TDP a significant improvement. Based off the 28nm process NVIDIA now has moved a forward in their technology and with all these new advancements the cards should run cooler and faster than ever.
NVIDIA's GPU Boost
The motto for Big Green with the launch of the GTX 680, better, faster, and smoother. What does this mean, let's start off with NVIDIA's GPU Boost, like Intel's Turbo Boost Technology NVIDIA now kicks in the power for games that by default are set to their basic settings for compatibility and unlock their other features. GPU Boost is a combination hardware/software technology that is supported by GeForce GTX 680. GPU Boost works in the background, dynamically adjusting the GPU’s graphics clock speed automatically based on GPU operating conditions. Dedicated hardware circuitry constantly monitors GPU power consumption. GPU Boost automatically adjusts clocks to achieve the maximum possible clock speed while remaining within a predefined power target. This is very interesting stuff that NVIDIA is trying out here, we will have to see how real world results are.
In previous generation NVIDIA cards 3D-Vision Surround required 2 GTX cards to power and run this technology. With another tech advancement using the new GTX 680 the user now has the ability to have up to 4 Monitors in 3D-Vision on a single GTX 680. Just think of the possibilities, 3 monitors on the bottom all in a row and one above in the center to still see and use your desktop. This is a great step for NVIDIA and I can't wait to hook up my 27'' ASUS 3D 120Hz Monitor and see how well the new GTX 680 does on my humble single 3D monitor, and in truth only the very rich can afford to even begin to look into getting a multi-monitor setup that consists of 3D Monitors, as that is a very costly investment.
Performance
