Nvidia GeForce GTX 680 review
Hugely fast, fairly power-efficient and not too expensive - a superb high-end graphics card
The GTX 680 is the first card we’ve seen with Nvidia’s new 28nm “Kepler” architecture. Kepler brings some notable advancements over previous Nvidia cards, notably the ability to run games on three monitors from one card in Surround mode and dynamic GPU overclocking.
Nvidia claims the GTX 680 is the “Fastest, most efficient GPU ever built”, and it certainly has some impressive specifications. There are 1,536 CUDA cores, a 1.01GHz core clock speed and the card’s 2GB of GDDR5 memory runs at a huge 1.5GHz. It’s a fairly compact card, too, at just 252mm long. This compares well to the AMD Radeon HD 7970 and its 277mm length, and you should have no problem fitting it in a medium-sized case. The card needs two six-pin PCI Express power connectors, and Nvidia recommends a 550-Watt power supply – this is no power-hungry monster like the dual-GPU GTX 590.
Despite its relatively modest size and power requirements, this is an extremely powerful graphics card. Its performance is helped by what Nvidia calls GPU Boost. This dynamically increases the GPU’s clock speed in 13MHz increments from its stock 1.01GHz up to a maximum of 1.058MHz where there is sufficient power and thermal headroom. This helps to overcome bottlenecks when not all of the card’s cores are being used to their full capacity, in a similar way to how Turbo Boost works on Intel processors when they’re running lightly-threaded applications.
The card was extremely fast in our benchmarks. Dirt 3, running at 1,920 x 1,080 with 4x anti-aliasing and Ultra detail, ran at an amazing 114fps. This is far faster than the Radeon HD 7970, which only managed 88fps. The only card we’ve seen with a better score in this test was the dual-GPU GeForce GTX 590, which is significantly more expensive. In Crysis 2, at 1,920 x 1,080 and Ultra detail, we saw a smooth 46.1fps – slightly faster than the HD 7970’s 42.9fps and again only beaten by the GeForce GTX 590’s 48fps.
While previous Nvidia cards needed to be paired with another GPU in SLI mode to run three monitors in Surround mode (Nvidia’s Eyefinity equivalent) the GTX 680 can run four monitors from one card, three of which can be used for gaming and one for other applications such as a web browser or chat program. It’s more versatile than Eyefinity; the GTX 680 has HDMI, DisplayPort and twin DVI outputs, and you can use any combination for Surround – we had no problems running Surround with one DVI, one HDMI and one DisplayPort monitor, for example.
Nvidia Surround and conclusion
Performance in Dirt 3 in Surround mode, running at 5,760 x 1,080, was also huge, with 49.6fps. This is significantly quicker than the AMD Radeon HD 7970’s 41.5fps when running at the same resolution in Eyefinity mode. Even during this punishing test the card became very warm but still stayed relatively quiet. However, we did find the GTX 680 slightly more power-hungry than AMD’s Radeon HD 7970; with the graphics card under load, our test PC drew 285W when fitted with the GTX 680, compared to 236W with the HD 7970. Also, when fitted with AMD’s card our test PC drew just 54W when Windows turned the monitor off, compared to 71W for Nvidia’s card – AMD seems to have the edge when it comes to shutting down as much of the graphics card as possible when it’s not needed.
Nvidia’s GeForce GTX 680 is a triumph; it’s the fastest single-GPU graphics card we’ve seen, is relatively compact and works with even mid-range power supplies. It’s also not particularly expensive; Nvidia’s suggested retail price is £429, which makes it only around £20 more than the Radeon HD 7970, and Nvidia’s card is significantly faster. It wins an Ultimate award.
Basic Specifications | |
---|---|
Price | £429 |
Rating | ***** |
Details | www.nvidia.com |
Award | Ultimate |
Interface | PCI Express x16 |
Crossfire/SLI | SLI |
Slots taken up | 2 |
Brand | nVidia |
Graphics Processor | Nvidia GeForce GTX 680 |
Memory | 2GB GDDR5 |
Memory interface | 256-bit |
GPU clock speed | 1.01GHz |
Memory speed | 1.50GHz |
Card length | 252mm |
Features | |
Architecture | 1,536 CUDA cores |
Anti aliasing | 8x |
Anisotropic filtering | 16x |
Connectors | |
DVI outputs | 2 |
VGA outputs | 0 |
S-video output | no |
S-Video input | no |
Composite outputs | no |
Composite inputs | no |
Component outputs | no |
HDMI outputs | 1 |
Power leads required | 2x 6-pin PCI Express |
Extras | |
Accessories | none |
Software included | none |
Buying Information | |
Warranty | one-year RTB |
Price | £429 |
Supplier | http://www.nvidia.com |
Details | www.nvidia.com |