Nvidia's Kepler architecture was introduced a year ago with the GeForce GTX 680, which felt reasonably comfortable as the best single GPU graphics card on the market. AMD had to cut prices and bring a special HD 7970 GHz Edition card to market to fill the value gap. Despite beating their rival, many believe Nvidia planned to make its flagship 600 series even faster with the GK110 chip, but deliberately held back with the GK104 to save money as it was in terms of the Performance was competitive enough.
That's not to say that people were necessarily disappointed with the GTX 680. The 28 nm part packs 3540 million transistors into a tiny 294 mm2 chip and delivers 18.74 gigaflops per watt with a memory bandwidth of 192.2 GB / s, while it triples the CUDA of the GTX 580 cores and doubled its TAUs – no small matter to be sure. Even so, we all knew the GK110 existed, and we were excited to see how Nvidia brought it to the consumer market – assuming it actually chose to. Fortunately, the wait is over now.
After the GTX 680 wore the single GPU performance crown for 12 months, it was dethroned by the new GTX Titan. Announced on February 21, Titan features a GK110 GPU with a transistor count that has more than doubled from 3.5 billion GTX 680 to a staggering 7.1 billion. The part has roughly 25% to 50% more resources than Nvidia's previous flagship, including 2688 stream processors (up 75%), 224 texture units (also up 75%), and 48 raster operations (a healthy 50% increase) ).
If you're curious, it's worth noting that there is "only" an estimated 25% to 50% performance gain as the Titan is clocked lower than the GTX 680. Given these expectations, it would be fair to assume the Titan would get the price be about 50%, which is about $ 700. But the Titan's prices aren't fair – and it doesn't have to be. Nvidia markets the card as a hyper-fast solution for extreme gamers with deep pockets and sets the MSRP at a whopping $ 1,000.
That puts the Titan in the field of the dual GPU GTX 690, or about 120% more than the GTX 680. In other words, the Titan won't be a good value in terms of price and performance, but Nvidia is undoubtedly aware of that This, and to a certain extent, we have to respect it as a niche luxury product. With that in mind, let's raise the Titan's hood and see what makes it tick before we run it through our usual benchmarks, which now include measurements of frame latency – more on that in a moment.
Titan's GK110 GPU in detail
The GeForce Titan is a real processing horse. The GK110 chip has 14 SMX units with 2688 CUDA cores and a peak performance of up to 4.5 teraflops.
As mentioned earlier, the Titan has a core configuration that consists of 2688 SPUs, 224 TAUs, and 48 ROPs. The card's memory subsystem consists of six 64-bit memory controllers (384-bit) with 6 GB of GDDR5 memory at 6008 MHz, which corresponds to a peak bandwidth of 288.4 GB / s – 50% more than the GTX 680.
The Titan we have is equipped with Samsung K4G20325FD-FC03 GDDR5 memory chips designed for 1500 MHz – just like the reference GTX 690.
Where the Titan lags behind the GTX 680 is its core clock rate, which is set to 836 MHz versus 1006 MHz. This 17% difference is easily offset by Boost Clock, Nvidia's dynamic frequency function, which can bring the Titan to 876 MHz.
The GTX Titan comes standard with a pair of dual-link DVI connectors, a single HDMI connector, and a DisplayPort 1.2 connector. There is support for monitors with 4K resolution, and up to four monitor screens can also be supported.