Introducing the GeForce GTX TITAN


In the Oak Ridge National Laboratory, in the U.S. state of Tennessee, a supercomputer ticks away. Officially the world’s fastest, this supercomputer is capable of performing an astonishing 27 quadrillion floating-point operations per second, 11 quadrillion more than the previous record holder. To emphasize this unmatched level of performance, its creators named it “Titan”, and it is this same name that we are bestowing upon our new flagship graphics card, the GeForce GTX TITAN.

Introducing the NVIDIA GeForce GTX TITAN.

Using a GPU identical in every way to the 18,688 designed for Oak Ridge’s record-breaking supercomputer, the GeForce GTX TITAN pushes the envelope of single-GPU performance by performing an unprecedented 4.5 trillion single precision floating-point operations per second. On average, TITAN is 35% faster than our flagship single-GPU graphics card, the GeForce GTX 680, but as our benchmark data shows game performance can increase by up to 85% at 2560x1600.

Available February 25th at a recommended retail price of $999.99, the GeForce GTX TITAN is cutting edge tech through-and-through, beginning with the new GK110 Kepler GPU. With 75% more CUDA Cores, an extra 7 Streaming Multiprocessors (SMX), and an extra SMX per Graphics Processing Cluster (GPC), GK110 is significantly more advanced than GK104; the GPU used in our fastest GTX 600 Series graphics cards.

The GeForce GTX TITAN GPU is comprised of 7.1 billion transistors. In comparison, the GTX 680’s GPU is comprised of ‘only’ 3.54 billion transistors.

For Surround users memory bandwidth has been doubled, and the amount of VRAM tripled. And for students, researchers and engineers there are 64 double-precision cores, giving them access to 1.3 teraflops of double-precision performance that could only be found previously in Tesla-equipped workstations and supercomputers.

Specifications GeForce GTX TITAN GeForce GTX 690 GeForce GTX 680 GeForce GTX 670 GeForce GTX 660 Ti GeForce GTX 660 GeForce GTX 650 Ti GeForce GTX 650
Chip GK110 2 x GK104 GK104 GK104 GK104 GK106 GK106 GK107
CUDA Cores 2688 3072 1536 1344 1344 960 768 384
Base Clock 837 MHz 915 MHz 1006 MHz 915 MHz 915 MHz 980 MHz 928 MHz 1058 MHz
Boost Clock 876 MHz 1019 MHz 1058 MHz 980 MHz 980 MHz 1033 MHz N/A N/A
Memory Configuration 6 GB 4 GB 2 GB 2 GB 2 GB 2 GB 1 GB 1 GB
Memory Speed 6.0 Gbps 6.0 Gbps 6.0 Gbps 6.0 Gbps 6.0 Gbps 6.0 Gbps 5.4 Gbps 5.0 Gbps
Memory Bandwidth 288 GB/s 384 GB/s 192 GB/s 192 GB/s 144 GB/s 144 GB/s 86.4 GB/s 80 GB/s
Power Connectors 1 x 6-pin
1 x 8-pin
2 x 8-pin 2 x 6-pin 2 x 6-pin 2 x 6-pin 6-pin 6-pin 6-pin
Outputs DL-DVI-I
TDP 250 W 300 W 195 W 170 W 150 W 150 W 110 W 64 W
SLI Options 3-way Quad 3-way 3-way 3-way 2-way N/A N/A
Price $999 $999 $499 $399 $299 $229 $149 $109

Next-Generation Construction


The dual-GPU GeForce GTX 690 is a thing of beauty and an engineering marvel, running quieter and cooler than a single-GPU GTX 680. Its precision crafted design is key to this feat, and so it was logical to reuse elements of its award-winning, cutting-edge cooling system for the GeForce GTX TITAN.

Starting with the GeForce GTX Titan’s Printed Circuit Board (PCB), we looked for ways to improve upon the GTX 690’s award-winning design. To that end, there’s an aluminum baseplate surrounding the GK110 GPU, drawing heat away it, other board components, and the six 500MB VRAM chips (a further six are on the rear of the card, bringing the total amount of VRAM to 6GB). On the GK110 GPU itself, we’ve applied a new thermal interface material from renowned Japanese firm Shin-Etsu, which is twice as effective as the thermal grease applied to GTX 680 GPUs, enabling the GTX Titan to run at higher clock speeds.

Design improvements enable the GTX TITAN’s Base Clock speed to be set 105MHz higher than the Base Clock speed of the Tesla K20X graphics cards in Oak Ridge’s supercomputer.

To collect the heat emitted by the GPU and VRAM, a high-performance copper vapor chamber is placed in contact with the GPU. Like the GTX 690’s vapor chamber, its process of controlled evaporation cools the card faster and to a better degree than any conventional heatsink and heatpipe combo. This vapor-chambered heat is then dissipated by a large, dual-slot aluminum heatsink that’s equipped with an extended fin stack that covers the entire area occupied by the GPU and VRAM. This increases the cooling area and overall performance of the cooling system.

Finally, to remove the collected and dissipated heat, a rear-mounted fan constructed from an acoustic-dampening material blows it out through the exhaust and away from the card’s components. In the GTX 690, this fan was mounted centrally, pushing heat out into the chassis, increasing overall system temperatures. This was necessitated by the front and rear placement of the GPUs, and the need to remove the considerable amount of heat as quickly as possible. With only one forward-mounted GPU, lower power consumption, and the aforementioned improvements, the GTX TITAN was able to utilize a traditional rear-mounted fan, reducing temperatures and enabling faster clock speeds.

As with the GTX 690, the TITAN’s exterior casing is made of cast aluminum, emphasizing the card’s cutting-edge design and use of premium-grade materials. The clear polycarbonate window also returns, allowing owners to marvel at the nickel-plated heatsink fins.

On the exterior edge, the GeForce GTX logo is illuminated by LEDs. Using third-party tools, their intensity can be manually adjusted, or be set to brighten and dim automatically with the level of GPU utilization, adding a neat effect to windowed cases.

Combined, these cooling features and design enhancements make the GeForce GTX Titan the quietest, coolest, high-performance graphics card ever produced; a claim corroborated by our data below.

In our tests, a 2-Way SLI GTX TITAN setup was 0.2 decibels quieter than a single GTX 690, and 2.7 decibels quieter than a single GTX 680.

GPU Boost 2.0: The Temperature’s The Limit

Before the launch of GPU Boost alongside the GTX 680 last March, video card clock speeds were determined by the extreme power usage of synthetic benchmarks. In games, GPU power consumption rarely ever hits such highs. Because GPUs lacked the means to monitor power consumption, clock speeds were conservative, lest a GPU go over the power limit and die an untimely death.

With the software and hardware innovations brought to the table by the Kepler GTX 600 Series, every application now runs at a guaranteed Base Clock speed, and if there’s extra power available, a Boost Clock is enabled, increasing clock speeds until the graphics card hits its predetermined Power Target (170 Watts on the GTX 680, for example). This dynamic clock speed adjustment is controlled by GPU Boost, which monitors a raft of data and makes realtime changes to speeds and voltages several times per second, ensuring the maximum level of performance possible in each and every application.

However, after careful evaluation of GPU Boost over the past year, NVIDIA’s engineers have now determined that GPU temperature is more commonly an inhibitor to performance than power consumption. As a result, we have developed GPU Boost 2.0, which increases the Boost Clock until the GTX TITAN hits the predetermined 80C Temperature Target. With this change, performance is increased by 3-7% over GPU Boost 1.0 at stock speeds, and by a larger degree on factory-overclocked models sold by our partners.

GPU Boost 2.0 increases Boost Clock performance by an extra 3-7% on our reference-design GTX TITAN.

In addition to switching from a power-based boost target to a temperature-based target, we’re also providing end users with more controls for tweaking GPU Boost behavior. Fan speed at GPU Boost 2.0’s 80C default too loud? Decrease the Temperature Target with third-party software and observe an instant reduction in fan speed. Want max performance regardless of noise? Crank the Target beyond 80C.

With a watercooling setup temperatures become irrelevant, allowing GPU Boost to maximize voltage and power usage, in turn cranking the Boost Clock way beyond the norm, wringing every last drop of performance from the Titan.

Overvolting: Step Into The Danger Zone

GeForce GTX TITAN owners who wish to push their cards to the limit will enjoy the new GPU Boost 2.0 controls discussed above, but for the most enthusiastic enthusiast we have one final addition of note: overvolting controls. Hidden away on GTX 600 Series cards, overvolting controls allow users to push power usage and performance beyond the safe limits guaranteed by manufacturers. Misuse can irreparably damage GPUs.

Through the aforementioned, third-party GPU Boost 2.0 control panels, GTX TITAN owners can enable overvolting once they have read and accepted warnings and damage waivers. Each individual card manufacturer may limit the degree of overvolting, or disable it altogether in the VBIOS.

Please be aware that overvolting may void the graphics card’s warranty, and that continued overvolting may cause the GPU’s silicon to degrade at an advanced rate, ultimately causing the graphics card to die a premature death through electromigration. As such, we don’t recommend the use of overvolting, but at the same time we don’t wish to put artificial limitations on a premium product designed for enthusiasts.

Display Overclocking: Throwing Off The Shackles

Many PC gamers play with VSync enabled to avoid unsightly image tearing that occurs when the frame rate is higher than the monitor’s refresh rate. Most are locked to 60Hz, limiting the max frame rate to 60 frames per second.

Unhappy with this stall in technological advancement, our engineers looked inside today’s LCD and LED-backlit screens, and discovered that many support higher refresh rates. For that reason, we’ve developed Display Overclocking, which allows GeForce GTX TITAN owners to quickly and easily overclock their refresh rates via the same third-party tools that control GPU Boost 2.0. And of course, it works with Adaptive VSync too, allowing games to be played at rates over 60Hz on a “60Hz” screen, without tearing, and without the stuttering associated with normal VSync.

Display Overclocking works with any panel operating at any speed, including the latest 120Hz screens.

GeForce GTX TITAN Performance: What You’re Itching To See

As we said in the introduction to this article, GeForce GTX TITAN is 35% faster on average than the GeForce GTX 680. With an extra 1,152 CUDA Cores, an extra 7 Streaming Multiprocessors, and 50% more memory bandwidth that’s hardly surprising. The data, however, shows that the performance gap is much wider at higher resolutions and with hardware anti-aliasing enabled.

At 1920x1080, the GTX TITAN didn’t break a sweat in our testing, hitting 60 frames per second in the newly-released, highly-demanding Crysis 3, with max game settings and 4xMSAA enabled. This level of performance was repeated and exceeded in other tests, too, with a top result of 154 frames per second being recorded in F1 2012. In Metro 2033, the modern-day system killer, the GTX TITAN powered to a respectable 50 frames per second, 59% faster than the GTX 680’s 31 frames per second.

On average, our stock-clocked GTX TITAN was 42% faster than the GTX 680, a figure that would increase dramatically with factory overclocks and higher Temperature Targets.

At 2650x1600, the GTX TITAN was finally able to flex its muscles, showing how its increased memory bandwidth can greatly aid frame rates when high levels of hardware anti-aliasing are enabled. In Crysis 3, the GTX Titan maintained a playable, smooth 40 frames per second, while the GTX 680 faded, dropping down to an unplayable 28 frames per second.

On average, frame rates at this highly demanding resolution were 48% faster than those of the GTX 680, a figure considerably higher than the 35% suggested by the specifications.

Combined with a powerful CPU, 2-Way and 3-Way GTX TITAN SLI setups will fly. To demonstrate, we teamed up with Vince "K|NGP|N" Lucido, the world's best overclocker, to break some records:

An air-cooled GTX Titan 3-Way SLI setup is capable of playing Crysis 3 at fluid frame rates at 5760x1080, using max game settings and 4xMSAA. LN2 not required.

Conclusion: A New Champion Is Crowned

The GeForce GTX TITAN is our fastest single-GPU graphics card, and also our most advanced, having been built from the ground-up with the latest technologies, techniques, and materials.

Its construction, performance, and efficiency will shape the next generation of consumer-level graphics cards. Its innovative GPU Boost 2.0 technology will be emulated by others. Its size and unprecedented acoustic and thermal performance will breed a new range of compact, supercomputer-class gaming PCs, beginning with Falcon Northwest’s Tiki. And its 4.5 teraflops of power will enable game developers to bring their titles to life like never before.

The GeForce GTX TITAN represents the ultimate in gaming from NVIDIA, built for enthusiasts who demand the very best, and for gamers who want a graphics card that will last for generations.