Fastest single videocard you can buy; GPU Tweak makes overclocking easy; quieter and more power-efficient than SLI.
Front-and-back airflow is hard to cool; gets flaky at high overclocks; $50 more expensive than competing cards.
Hello, gorgeous. That’s what we said when we first laid eyes on Nvidia’s reference design for the GeForce GTX 690 , which combines two full 28nm GK104 GPUs into one PCB and covers them with the best-looking cooling shroud we’ve seen on any videocard. Our in-depth analysis of the reference card can be found here , but we can’t verdictize a reference card. If you’re wondering how this Asus GTX 690 differs from the reference card Nvidia sent us, wonder no more: It’s exactly the same, except the edges of the PCB are a slightly different color.
Still gorgeous. Still massive.
Like the reference card, the Asus GTX 690 has a whopping 3,072 CUDA cores, 256 texture units, 64 ROPs, and 4GB of memory at 6,008MHz combined speed. Unlike previous dual-GPU cards, it has all the cores and texture units intact on each GPU, and each GPU is only slightly underclocked, to 915MHz with a 1,019MHz boost clock. The GTX 680 has a base clock of 1,006MHz and a boost clock of 1,058MHz.
Since this is essentially a duplicate of the reference design, it carries the same 48-lane PLX bridge chip connecting the GPUs to each other and to the x16 PCIe 3.0 interface. It also uses the reference cooling shroud, for which we can’t fault Asus one bit. Why mess with a great thing?
The cooling shroud is chromium-plated cast aluminum with a magnesium-alloy fan housing in the center of the card. The fan takes in air centrally and blows it into vapor chambers atop each GPU, venting the hot air toward both the front and back of the card—not ideal for standard front-to-back airflow. The vapor chambers over the GPUs are covered with polycarbonate windows. The fan is fairly quiet even when running at full load. It’s not silent, but it’s less noise than a pair of cards would be.
The Asus GTX 690 has a TDP of just 300W (a single GTX 680 has a TDP of 195W) and requires two 8-pin PCIe power connectors. Its rear panel contains two dual-link DVI-I ports, one dual-link DVI-D, and a Mini DisplayPort 1.2 connector. The card, like the reference board, is 11 inches long.
Asus ships the GTX 690 with its GPU Tweak software, which allows for easy overclocking of the memory clock and GPU boost clock, as well as voltage tweaks. We were able to overclock the boost clock to around 1,100MHz without sacrificing stability, but the card got flaky as we went higher. More tweaking would doubtless have given us a long-term stable overclock.
In our benchmarks, the performance of the Asus GTX 690 was indistinguishable from the reference card—that is, it was within a few percentage points of a set of stock-clocked GTX 680s in SLI, while consuming less power. Depending on the benchmark, it’s anywhere from 46 percent to 95 percent faster than a single stock-clocked GTX 680. On all but three benchmarks, it beat a pair of reference Radeon HD 7970s—we’ll test it again against GHz edition 7970s when we track down a pair of them.
Very few people need the power of two top-of-the-line GPUs. But if you do, it’s better to have them on one videocard rather than two: It’ll draw less power, be easier to cool, and will give you the option of adding another GTX 690 down the line. We’re still wild about the looks of the cooling shroud, if not about the fact that it exhausts hot air in both directions. So far the only GTX 690s we’ve seen on the market have been stock-clocked ones using the reference shroud, and AMD’s dual-Radeon card is nowhere to be found, so a GTX 690 is your only dual-GPU option right now. Asus’s is slightly more expensive than, say, EVGA’s, and it’s not clear what you get for the extra $50. But at that point, you’re already spending a grand, so does the extra $50 really matter?
|Asus GTX 690 ||Nvidia GTX 690 Reference||Nvidia GTX 680 SLI||AMD Radeon HD 7970 CrossFireX||EVGA GTX 680|
|3DMark 11 Perf||P15,056||P15,104||P15,804||P13,817||P9,555|
|3DMark 11 Extreme||X5832||X5800||X6072||X5352||X3249|
|3DMark Vantage Perf||P44,501||P44,501||P45,205||P44,180||P34,339|
|Unigine Heaven 2.5 (fps)||59.8||59.8||60.8||57.6||31|
|Shogun 2 (fps)||35.4||37.9||38||75||18.7|
|Far Cry 2 / Long (fps)||183.7||184.7||187.9||186.3||107.3|
|Dirt 3 (fps)||123||122.9||124.5||114.5||73|
|HAWX 2 (fps)||WNR||224||225||229||131|
|Metro 2033 (fps)||29||29.6||29.5||29||16.3|
|STALKER: CoP DX11 (fps)||64.2||64.6||66.1||76.7||34.3|
|Just Cause 2 (fps)||80||79.31||81.03||91.85||54.7|
|Batman: Arkham City (fps)||103||102||104||86||58|
|Base Clock (Actual)||915MHz||915MHz||1,006MHz||1,000MHz||1,006MHz|
Best scores bolded. Our GPU test bed consists of a stock-clocked Intel Core i7-3960X on an Asus P9X79 Deluxe board with 16GB DDR3/1600, a 256GB Samsung 830 Series boot SSD, and a 1,050W Thermaltake Toughpower Grand PSU, in a Cosmos II chassis. All tests performed at 2560x1600 with all settings maxed and 4x MSAA except where noted. Power use measured with a Watts Up Pro.