It’s a small class of PC enthusiasts who will spend US$500 on a graphics card. It takes pretty demanding tasks — perhaps a large high-res monitor or two or a taste for extreme image quality — to make use of that sort of performance. Most modern games run well enough on a graphics card costing half that price. If you’re one of those people for whom “well enough” isn’t enough, then the GeForce GTX 680 is for you. Among expensive graphics cards, it takes the crown for delivering the best performance, excellent features, and very impressive energy efficiency.
With the GeForce GTX 680, Nvidia introduces a new graphics chip with a new architecture, code-named Kepler. It’s actually similar to the Fermi architecture found in the GeForce GTX 580 and the GTX 480 before it. Nvidia has taken the fundamental principles of the Fermi architecture and re-made it with an aim at greater efficiency; improving performance per watt and performance per square millimeter. The Fermi chips are also manufactured on a smaller 28 nanometer process, where the GTX 580 was made on a 40nm process. That means lower power draw, higher clock speeds, and a chip much more densely-packed with transistors. In fact, Kepler is considerably smaller than Fermi at less than 300 mm2 (full-size Fermi chips are 520mm2).
The new GPU reduces the amount of control logic in each cluster of stream processors, sacrificing some potential performance in GPU-based general computation for better performance in 3D graphics applications. The result is a chip with three times the stream processors of Fermi, twice the texture units, and a clock speed just over 1GHz. There are fewer render back-ends and a narrower memory bus, but those are each offset by the higher clock speeds. The result is a US$500 high-end graphics card that delivers industry-leading graphics performance while drawing very little power (for a card in this price range). It’s so energy efficient that the standard board design features two six-pin PCIe power plugs, where most US$500 graphics cards require one six-pin and one eight-pin to supply enough power.
Nvidia is introducing a handful of new technologies with the 600 series GeForce cards, too. GPU Boost is very similar to the Turbo Boost and Turbo Core technologies you find in Intel and AMD processors — if the card is running fairly cool and not drawing too much power, it automatically boosts the clock speed for better performance. AMD does this with its graphics cards, so it’s nice to see Nvidia adopt the concept. A new form of anti-aliasing, called TXAA, should help provide smoother edges with less of a hit to performance than other forms of anti-aliasing. Our tests show it to be a very good tradeoff between performance and quality. Adaptive vsync is another nice tradeoff between the smoother performance of having vertical synchronization disabled and the artifact-free gaming of enabling vsync.
Of course, if you’re in the market for a US$500 graphics card, you’re really concerned with how fast it is. The good news is, all this focus on energy efficiency has not cost Nvidia the performance crown. The GeForce GTX 680 manages to outpace its obvious competitor, AMD’s similarly-priced Radeon HD 7970, in most of our tests. The improved tessellation performance in Nvidia’s new chip really help it shine in synthetic tests like Unigine Heaven. It cranks out a very impressive score in 3DMark 11, too.
When we average the performance of the two cards among four popular, modern, demanding games (Crysis 2, Just Cause 2, Dirt 3, and Metro 2033), we see the GTX 680 just barely edging out the Radeon HD 7970. In some games it’s a bit faster, in some a bit slower, but the performance is generally a little higher overall. This is with a graphics card that is slightly cheaper (around US$20-30 less if you shop around online) and uses a bit less power.
At the end of last year, we praised the Radeon HD 7970 from AMD for its extremely high performance and fantastic energy efficiency. In the months since then, AMD has improved performance with driver updates, and yet Nvidia’s new GeForce GTX 680 still manages to beat it on both counts, typically drawing less power and delivering slightly faster performance. Together with a couple of nice new features like TXAA and Adaptive vsync, it’s a clear winner. AMD’s 7000 series relies on a graphics architecture that is drastically different from previous AMD chips, so we may still see much better performance with driver improvements, but right now, if you have US$500 to spend on a graphics card, the GeForce GTX 680 is the clear winner.