NVIDIA GeForce GTX Titan Review

ccokeman - 2013-02-16 20:15:09 in Video Cards
Category: Video Cards
Reviewed by: ccokeman   
Reviewed on: February 19, 2013

NVIDIA GeForce GTX Titan Introduction:

It's been just shy of a year since NVIDIA debuted the Kepler architecture with the launch of the GTX 680, a card that ratcheted up the heat on AMD to deliver something that could outshine the beast that was Kepler. This weeks news that AMD is going to keep riding the wave of its HD 79XX series for the rest of the year puts NVIDIA in the driver's seat for now with the introduction of the GeForce GTX Titan. However, what we were treated wih the introduction of Kepler was a cut down version of the architecture in the form of GK104 instead of the full implementation GK110. Much the same story line we had with the Fermi-based GTX 480 then a year later the GTX 580 making for a comparable timeline and strategy between the two generations. Each time the cards were delivered late in the game with AMD gaining market share only to be outperformed once NVIDIA delivered its card to the market.

GTX Titan is named after the world's fastest super computer at Oak Ridge National Laboratory, aptly named Titan. The Titan Super Computer is home to 18,668 NVIDIA Tesla K20X GPU accelerators built using the GK110 Kepler core to deliver 17.59 petaflop/sec in the LINPACK benchmark. Tesla K20X's compute performance firmly entrenched allowed for the GK110 architecture to filter down to the consumer level with NVIDIA's GeForce GTX Titan. The hallmarks of this card are a new set of features including GPU Boost 2.0, Overvoltage controls, display overclocking, new vapor chamber cooling solution, small form factor PC compatibility, Tri-SLI, improved double precision compute performance, and improved fan acoustics.

Basic specifications for GTX Titan include five GPC, 14 SMX, 2688 CUDA Cores (Single Precision), 896 CUDA Cores (Double Precision), 448 texture units, 48 ROPs, 6GB of GDDR5 memory though a wider 384-bit bus, 28nm process, and finally 7.1 billion transistors with a 1536K L2 cache. Base clock speed on the CUDA cores comes in at 837MHz with a boost clock of 876MHz while the 6GB frame buffer is clocked at 6008MHz (1502MHz effective). Just the specifications alone show the performance potential that Titan should deliver on the gaming front. However, for this introduction to Titan we'll focus on the card itself and what the feature set will do for the end user with benchmarks to follow up in another article shortly. So let's dig into Titan.

NVIDIA GeForce GTX Titan Closer Look:

First off, we can take a look at the card to see just where NVIDIA put the performance. The card I have is a pure reference card. If you have seen a GTX 690 you get an idea where the styling cues come from. As much as the GTX 690 was a thing of beauty for a video card, the GTX Titan follows in the same mold with an aluminum frame and clear polycarbonate window over the vapor chamber-based cooling solution. At 10.5 inches long, the GTX Titan is able to fit into more small form factor chassis to deliver world class gaming performance without the need to have at least a mid-tower chassis. Using a blower style fan, it's able to push all the thermal load out of the chassis rather than heat up the other internals. The back side of the PCB shows that half the 6GB frame buffer is populated here but without any additional cooling for the modules. The GeForce GTX logo on the top of the card lights up to show just what kind of graphics horsepower you are running, functions as a power indicator, can be manually adjusted with board partner tools, and can also function as a graphics load indicator. The GTX Titan is PCIe 3.0 compliant with backwards compatibility built in.

 

 

 

 

 

 

 

Display connectivity does not change from earlier Kepler-based video cards with a pair of Dual Link DVI ports, a single full size DisplayPort 1.2 port, and an HDMI 1.4a port. This configuration supports up to four monitors in a 3+1 surround configuration. As a card built to run cool and quiet with an expressed goal of running in small form factor PCs, the airflow pushed through the GTX Titan is pushed outside the chassis to keep from heating up the PC components in an SFF build. It's functionality that the GTX 690 cannot provide. A series of openings on the mounting plate seemingly provide for the airflow requirements without impacting noise levels so that the thermal load is pushed completely out of the chassis. The back end of the card is open to draw air in through a fin array starting the cooling process early in the cycle, most likely cooling the VRM components.

 

 

As a single GPU card, the GTX Titan supports up to a four-way SLI solution to offer the highest possible gaming performance/experience. It's being touted as the only GPU solution that can run Crysis 3 maxed out at playable frame rates. The efficient dual slot cooling allows the cards to fit into a board that supports a four-way solution. A pair of SLI bridge connections are found in the usual spot. The GTX Titan will need a power supply of at least 600 watts and 6-pin and 8-pin PCIe power connectors. NVIDIA states the GTX Titan has a max board TDP of 250 watts with a 100% power target and 265 watts with a 106% maximum power target, which makes sense based on the power inputs to the GTX Titan.

 

 

NVIDIA has made great strides in reducing the thermals and noise signature of its Kepler-based products. The GTX Titan is no exception to this trend. Titan is equipped with a new copper vapor chamber-equipped cooling solution; a proven solution in the past that uses an extended fin design to deliver the best cooling solution possible in the footprint available. An aluminum base is used to cool the 6+2 phase VRM and DRAM modules on the PCB. This base assembly functions not only as a heat sink but as added structural rigidity to keep the PCB from flexing. Blower-style cooling solutions using a squirrel cage style fan have gotten a bad rap in the past for good reasons. NVIDIA has solved these issues on both the GTX 690 and the product most like Titan, the GTX 680. That continues here with added enhancements that include acoustic dampening materials on the fan as well as improved fan control algorithms that provide the optimum voltage to control the fan speed, noise, and temperature. A temperature target of 80 °C is the intended range before the fan speed really starts ramping up the duty cycle.

 

NVIDIA's PCB on the GTX Titan differs from its K20 cousins and is designed for use in a high performance gaming computer. To that end NVIDIA's engineers provided a 6+2 phase power circuit with six phases feeding the GK110 core with the additional two phases feeding the 6GB of GDDR5 memory. The power design supports overvoltage controls that can be taken advantage of with its GPU Boost 2.0 feature set.

 

The Kepler core used in the GTX Titan is built on a 28nm process that houses 7.1 billion transistors. It is the full implementation of the Kepler architecture first seen in March of 2011 as GK104 on the GTX 680. This time we get five Graphics Processing Clusters, 14 SMX with 192 CUDA cores that make up the 2688 core count, 1.5MB of shared L2 cache, 224 Texture units, 48 ROPs, and a full 6GB of GDDR5 memory running on a new-for-NVIDIA 6x64-bit (384-bit) bus. A total boost of 75% over the GTX 680's core, texture, and ROP count with a massive jump in frame buffer size. GPU Boost 2.0 is an upgrade over the prior versions that dynamically sets the clock speeds of the GPU core to the max possible clock based not only on the power threshold but more importantly the thermal threshold of 80 °C. By tweaking the power profile you can improve performance with your card. Samsung is the memory supplier of choice for the GTX Titan with its 6GB of memory to manage. Clock speeds are dynamic on the GPU cores due to how GPU Boost 2.0 operates, but a base core clock of 837MHz on the 2688 CUDA cores is the lowest you will see on the core with 1502MHz on the GDDR5 memory. The GTX Titan is more than a one trick pony however, as it's capable of running in double precision mode with 896 CUDA cores active to deliver over one teraflop of double precision performance. The clock speed is reduced but provides a pathway to bring developers into the fold with an easy entry point into CUDA programming.

 

 

Just on paper the GTX Titan, with its massive jump in core and memory count, is that as a single GPU gaming is no where near finished. The specifications flat out rock and should provide an interesting performance head to head with the GTX 690 and a 6GB GDDR5-equipped HD 7970. Now that we have seen that the GTX Titan is a great looking card in the vein of the GTX 690, we can look at the feature set that the GTX Titan comes equipped with.

NVIDIA GeForce GTX Titan Closer Look:

Architecturally the GTX Titan is using the by now familiar Kepler architecture first shown when the GTX 680 was released just shy of a year ago. This time however, we get the full GK110 version of the silicon. Packed full of 7.1 billion transistors using a 28nm process, the GTX Titan is equipped with five Graphics Processing Clusters, 14 streaming multiprocessors each with 192 CUDA cores (2688 total), a total of 224 texture units, 48 ROPs, and an impressive 6GB frame buffer running through a 6 x 64 (384) bit bus to improve high resolution gaming not just now but into the future. Clock speeds for the GTX Titan are 837MHz on the core with a GPU Boost 2.0 clock of 876MHz as long as the core runs within both temperature and power limits. The 6GB of GDDR5 comes with a 6008MHz (4 x 1502MHz) clock speed for improved memory bandwidth. When you get to the end of the rainbow you are left with the supposed most powerful single GPU card currently in production.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Each new iteration of an architecture is going to come with an enhancement to the feature set. The GTX Titan is no exception to that rule. To start the show off we have GPU Boost 2.0. With GPU Boost 1.0 the ultimate goal was to reach as high a clock speed as the GPU could run while remaining below a specific power threshold regardless of temperature implications. This led to limits in the performance when temperatures were kept low through water cooling or a more efficient air cooling solution. NVIDIA realized that a change was needed as silicon degradation happened not only from the application of too high a voltage to the core, but from too high a temperature as well. GPU Boost 2.0 works on a thermal target instead of a temperature target allowing the clock speed to boost as high as it can while the GPU Core is at or under the temperature target of 80 °C. By dynamically adjusting the voltage, fan speeds, and GPU clocks, the GTX Titan can deliver the highest possible performance with the lowest possible noise profile all while staying at or under the 80 °C target temperature. To increase the maximum boost clock potential you can simply adjust the temperature target higher. By increasing the temperature threshold, the core can set higher clock speeds and again manage the acoustics to keep the card no higher than the temperature target. The charts below show this in a graphic that is easy to understand.

 

 

 

GPU Boost 2.0 provides an incredibly complex yet simplistic approach to giving the gamer/end user the best possible experience by allowing the operating parameters be set. Cooler temperatures, higher performance through higher clock speeds, or a lower noise solution that will fit right at home in a small form factor PC. The GTX Titan and GPU Boost 2.0 makes it possible to have the fastest single GPU in a small form factor chassis. It's a product that has been next to impossible to build until now.

 

 

If the performance that GPU Boost allows is not enough, then NVIDIA does support overvolting the GTX Titan as a means to further boost performance. By allowing the voltage to be increased above what it considers the Vrel, New Vrel, or voltage it considers safe for the silicon over a usage period via GPU Boost 2.0, performance can be impacted over time. That time period, as we know, can be short or a year from now. By increasing this to the MAX voltage, the user can push the performance targets even higher. Of course this comes with risks and to use this feature with the GTX Titan, the user must acknowledge the risks through a warning pop up.

 

 

One of the more interesting features is the ability to overclock or overdrive your display refresh rate to achieve a potentially higher FPS when running with VSYNC on. Typically you are limited to the refresh rate of the monitor. The monitor sends its capabilities to the display adapter through its Extended Display Identification Data or EDID. If your monitor can handle the increased refresh rate you stand to see an increase in the playable FPS in games. A win across the board in terms of gameplay. This tool is not directly available through NVIDA but will be supplied by the AIB partners.

 

NVIDIA GeForce GTX Titan Specifications:

Graphics Processing Clusters
5
Streaming Multiprocessors
14
CUDA Cores (single precision)
2688
CUDA Cores (double precision)
896
Texture Units
224
ROP Units
48
Base Clock
836 MHz
Boost Clock
876 MHz
Memory Clock (Data rate)
6008 MHz
L2 Cache Size
1536K
Total Video Memory
6144MB GDDR5
Memory Interface
384-bit
Total Memory Bandwidth
288.4 GB/s
Texture Filtering Rate (Bilinear)
187.5 GigaTexels/sec
Fabrication Process
28 nm
Transistor Count
7.1 Billion
Connectors
2 x Dual-Link DVI
1 x HDMI 1 x DisplayPort
Form Factor
Dual Slot
Power Connectors
One 8-pin and one 6-pin
Recommended Power Supply
600 Watts
Thermal Design Power (TDP)1
250 Watts
Thermal Threshold2
95° C

 

NVIDIA GeForce GTX Titan Features:

 

 

All information courtesy of NVIDIA

NVIDIA GeForce GTX Titan:

Released just before the launch of its new GPU, NVIDIA put together a Free-To-Play bundle for users purchasing an NVIDIA video card. Free-to-play has become a successful model to bring in a wide range of gamers. In this bundle are games that take advantage of all the NVIDIA ecosystem has to offer. Follow the link here for a quick look at the bundle and what you can expect when you purchase a new NVIDIA video card. Sadly it's a day or two early for all of the benchmarks that you are looking for, as today's look at NVIDIA's GTX Titan was a way to preview the card and show off its feature set, as well as taking a closer look at what is a truly amazing piece of graphics hardware.

With a new feature set that allows for a customized operating environment to bringing back meaningful Double Precision performance for the developer working their way into CUDA programming, it is a new way of going to business. From all the product specifications and features it looks as though the GTX Titan has the muscle to drive just about any game out at high end resolutions and settings. Check back for the performance benchmarks and final wrap up on Thursday! To see performance results read the review here.

 

Take a look at the video NVIDIA has put together to see just what the GTX Titan is all about.