Shown above is the reference GTX 680 card and without doubt it looks like an NVIDIA card, appearing more like a new revision GTX 570 than the GTX 580 which had visible heatpipes. Turning round to the back of the card we can see that NVIDIA use a black 10" PCB, that the memory is all on the front of the product and that two SLI connectors are present which lets us know we can connect three cards together for enhanced performance.
Underneath the black plastic shroud is a block of aluminium fins with the acoustically dampened fan directly beside them, passing air through and out of the bracket at the end of the card. Underneath these fins is a copper GPU block with three heatpipes and sitting below that cooler are eight GDDR5 memory chips as well as the 28nm Kepler/GF104 core containing 3.54 billion transistors.
Around at the card outputs we have four present on each GTX 680 with two Dual-Link DVIs starting us off, followed by a HDMI 1.4a and then a full size DisplayPort 1.2 connector. The GTX 680 is capable of running four screens at one time with three available for surround gaming at resolutions such as 5760×1080 and the fourth running off a separate controller for desktop applications such as messaging. It is also worth noting that NVIDIA now offer bezel correction hotkeys for quick in-game changes to show any hidden content and that when setting up three screens we simply click the enable option, follow the simple wizard which allows us to drag and drop the right order making the process easy. As an extra touch the taskbar is now present on the centre display and we can maximise windows to individual screens not just the entire bank of displays.
In terms of power requirements we have two 6-pin connectors which point out the top of the card rather than the back. The GTX 680 has a rated TDP of 195w and runs on a four phase power design for the GPU with two further phases for the memory.
The reference specifications for the GeForce GTX 680 are 1006MHz core (base clock) with a boost clock of 1058MHz and the 2GB of GDDR5 memory is set to 3004MHz. There are 128 Texture units within the core along with 32 ROPS, 4 graphics processing clusters and 8 streaming multiprocessors. For the CUDA Cores NVIDIA have gone for a 1536 "core" design and our memory bus is 256-bit (4×64). PCIe 3.0, DirectX 11.1 and DirectCompute are all supported on this card as is acceleration of high definition content and the card can output 7.1 audio over HDMI as well as support PhysX, 4K displays and 3D Vision.