Nvidia's long-awaited move to 28nm technology is finally upon us today with the launch of the GeForce GTX 680. Billed as the most efficient GPU ever put together, the GTX 680 has a very reasonable power envelope of just 195W, breaking with the tradition of every new high-end graphics card requiring more power than the previous generation. That power figure is important for another reason: Nvidia no longer has a static clock speed for its graphics card.
The GeForce GTX 680 has a 1,006MHz base clock and a 1,058MHz boost clock and will dynamically alternate between those values depending on load and power consumption. Of course, you can also break out of that range, with the card downclocking itself as low as 324MHz when idling and easily reaching stable speeds in excess of 1,100MHz on the upside. Nvidia expects its board-making partners to really go to town and overclock — which in this case means raising the operational range higher — the GTX 680 right out of the gate, while advanced tweaking software will be provided to let you make similar changes on the fly as well. The reference card comes with two 6-pin auxiliary power connectors, but the more extreme variants will probably be equipped with 8-pin power jacks to help you max out performance.
Unlike its nemesis, the AMD Radeon HD 7970, Nvidia's new card doesn't require any application-specific software profiles for power management; it's all done using hardware monitors, which track power, temperature, and utilization and react almost instantaneously. Clock speed adjustments will happen in 13MHz increments every hundred milliseconds, meaning that you'll end up with different clock speeds for different frames inside your game.
Other salient specs on the GTX 680 include 32 ROP units, 128 texture units, and 2GB of GDDR5 VRAM running at a screaming 6GHz effective data rate over a 256-bit memory interface. That makes for total memory bandwidith of over 192GBps. Nvidia has reengineered its CUDA processing cores to make them significantly smaller and now counts 1,536 inside its latest graphics chip, which is composed of 3.54 billion transistors, almost 800 million fewer than the HD 7970. On the stock card configuration, you'll find two dual-link DVI ports, one HDMI output, and one DisplayPort connection. Those can be used to power three-monitor gaming off just one GTX 680, including 3D Vision Surround support, plus the new card can support a fourth "companion" display for things like chat, email, or cheat sheets alongside the others.
Alongside its new flagship GPU, Nvidia is also introducing a pair of new antialiasing modes. The GeForce GTX 680 is obviously capable of multisampling antilaliasing (MSAA), but Nvidia is now adding a new shader-based FXAA, which can substitute MSAA and deliver the same results at much lower processing cost or work alongside it to really smooth out diagonal lines on screen. FXAA will work across Nvidia's product range. The company also showed off an extra-special TXAA, which looks even better, but won't be available yet, and when it does it's likely to be limited to the new Kepler architecture. In any case, color us impressed.
All the foregoing should give you plenty of reason to contemplate an upgrade, and that's without even factoring in subtler additions like a new integrated heatpipe and acoustic dampening materials to make the card run cooler and quieter and Adaptive Vsync to try and get you the cleanest possible image. To dive into the full detail of Nvidia's GeForce GTX 680, browse through the list of comprehensive reviews at the bottom of this article.
The Nvidia GeForce GTX 680 should be available immediately, priced at $499 in the US, £429 in the UK (including VAT) or €419 in Europe before taxes.