Event Coverage
Confab in Kuala Lumpur – NVIDIA Announces New High-Efficiency GPUs
By Rico Cruz - 22 Mar 2012,11:37pm

Meet Kepler

Faster, Smoother, Richer

For the past two years, NVIDIA's Fermi architecture has been at the forefront of NVIDIA's GPU strategy. The design undperpins the 400 and 500 series GPUs, which feature 512 NVIDIA CUDA cores matched with upwards of 1GB of GDDR5 memory. This combination allowed the cards to hit core clock speeds in excess of 800MHz (in stock) and memory speeds of over 4GHz.. These numbers represented a significant improvement over NVIDIA's last GPU architecture, the GeForce 8 series, thanks to major changes in the basic hardware design of the Fermi-based cards. All that power comes at a significant cost however, as the most recent flagship model, the GeForce GTX 580, sucks down over 240W of electrical power. Because of this, increased efficiency has become the major thrust of NVIDIA for 2012, as they begin the first quarter of the year with the new 600 series, which was announced to members of the Southeast Asian IT media in Kuala Lumpur, Malaysia.

Nick Stam, Director of Technical Marketing for NVIDIA, explains the ins and outs of the Kepler GPU architecture. 

With the 600 series, NVIDIA introduces its next-generation GPU architecture, called "Kepler". Kepler aims to continue NVIDIA's commitment to delivering a "Faster, Smoother, Richer" graphics experience without the need for large amounts of electrical power. "Faster" is achieved by upping the number of CUDA cores on the GPU, which has been increased dramatically   from 512 to 1536 (8 groups of 192 cores each). The increase in cores allows the 600 series to achieve core clock speeds of over 1GHz.

The new Kepler GPU architecture features 1536 cores, arranged in 8 groups of 192.

To demonstrate just how much of an improvement in performance this is, NVIDIA representatives ran a game graphics demo called "Samaritan." Last year, it took three NVIDIA GeForce GTX 580s to run it. This time, it took just one GeForce GTX 680 graphics card. Raising the number of cores also increases efficiency, and so the GeForce GTX 680 offers twice the performance per watt than the previous GeForce GTX 580. Using a GPU performance manager application called EVGA Precision, the NVIDIA presenters showed the graphics card get up to 1GHz, but maintain power consumption of just 195W (the GeForce GTX 580 uses 244W) , thanks to its dynamic power management. EVGA Precision is only one of the various power management applications that will be bundled with the GeForce GTX 680, depending on the vendor. These applications allow you to set performance and voltage caps in order to increase efficiency, or you could use them to overclock the card - up to 1200MHz according to NVIDIA.

According to NVIDIA, the GeForce GTX 680 and its new Polymorph 2.0 engine has twice the tessellation than its predecessor, and up to four times more than the AMD HD 7970. Better tessellation means more graphics triangles are rendered, translating to more detailed, and more realistic images.

For "Smoother" graphics, the NVIDIA 600 Series introduces TXAA and Adaptive V-Sync. TXAA stands for Temporal Anti-Aliasing, an Anti-Aliasing algorithm that samples not just individual pixels within a frame for smoother color and texture gradation, but also samples pixels from the frame before and the frame after, making for more realistic motion of materials. It bests previous algorithms used by NVIDIA such as FXAA and MSAA 8x. Adaptive V-Sync on the other hand, is NVIDIA's new technology for eliminating the problems caused by the difference between a game's frame rate and the monitor's refresh rate.

The GeForce GTX 680 can drive three 3D monitors plus one extra for browsing or chatting. NVIDIA calls this 3D Vision Surround. The previous generation GPUs could only drive a maximum of two 3D monitors.

When frame rates drop below the monitor's refresh rate, lag occurs. On the other hand, when frame rates go below the refresh rate of your monitor, page-tearing occurs. Adaptive V-Sync controls the shifts in frame rate so that these picture distortions are minimized. Lastly, for "richer" graphics, the new cards now have enough processing power to render difficult textures like fur more realistically than ever before. Physics effects get a boost as well, as the new GPU was demonstrated to have the ability to accurately simulate random dynamic fracturing of objects - something we have not seen in games yet.

NVIDIA GeForce GTX 680 Official Specifications

CUDA Cores 1536
Base Clock 1006MHz
Boost Clock 1058MHz
Memory Configuration 2GB/256-bit GDDR5
Memory Speed 6.0Gbps
Power Connectors 6-pin + 6-pin
TDP 195W
Outputs 2x DL-DVI, HDMI, DisplayPort 1.2
Bus Interface PCI Express 3.0