For competitive and highly motivated people, trials and hurdles are simply tests to benchmark their excellence. After a certain point, it is not even about the competition anymore but a test of willpower and a challenge to oneself. Failure is not an indication to stop but a learning tool. Had Thomas Edison stopped trying after the 12th attempt of figuring out which filament can withstand the heat inside an electric bulb, the world would be a darker place today. Considering current PC games are ports from 6 and 7-year old consoles, should video card manufacturers stop producing new hardware until software developers catch up?
PC technology had always been in the vanguard of gaming, accepting no compromises and always looking toward the horizon for the next challenge. Waiting for console developers to get their act together will only result in failure. In the history of personal gaming, consoles have never been more technologically advanced than PC gaming so there is no reason to follow Sony or Microsoft’s lead. As one of the companies at the fore, leading the charge of PC gaming progress, NVIDIA has set out to create the fastest single-GPU video card on the planet, not by simply boosting the clock speeds and reselling it at a higher price point, but by tailoring their latest Tesla GPUs for the desktop market.
The GeForce TITAN video card is meant to perform beyond any single GPU Kepler video card that has been previously released. The GeForce TITAN is equipped with the GK110 core, with five graphics processing clusters, 14 streaming multiprocessors, 224 texture units and 48 ROP units. The GeForce TITAN runs at an 836MHz Base clock with a Boost clock of 876MHz. In terms of memory, the GeForce Titan has 6GB of GDDR5 to utilize across a 384-bit memory interface providing a total memory bandwidth of 288.4 GB/s. Usually, this much performance is expected to be a power hog but, due to further optimizations on NVIDIA’s part, the GeForce GTX TITAN TDP is a scant 250W, requiring only a single 6-pin and 8-pin power connector.
|Graphics Processing Clusters||5|
|CUDA Cores (single precision)||2688|
|CUDA Cores (double precision)||896|
|Base Clock||836 MHz|
|Boost Clock||876 MHz|
|Memory Clock (Data Rate)||6008 MHz|
|L2 Cache Size||1536K|
|Total Video Memory||6144 MB GDDR5|
|Total Memory Bandwidth||288.4 GB/s|
|Texture Filtering Rate (Bilinear)||187.5 GigaTexels/sec|
|Transistor Count||7.1 Billion|
|Form Factor||Dual Slot|
|Power Connectors||One 8-pin and one 6-pin|
|Recommended Power Supply||600 Watts|
|Thermal Design Power (1)||250 Watts|
|Thermal Threshold (2)||95C|
1 TDP is a measure of maximum power draw over time in real world applications. It does not represent the maximum power draw in pathological cases such as Furmark. Please see http://en.wikipedia.org/wiki/Thermal_design_power for more information.
2 The GPU is designed to operate safely up to this temperature. If for any reason the GPU exceeds this temperature, the clock speed will automatically be dialed down.
To benchmark video cards, we will use our suite of popular games, as well as synthetic benchmarks. For testing purposes, all scores will be posted on a graphical chart and broken down into sections relative to their classifications (Synthetic, DirectX 11, etc). Graphs will measure Frames per Second (FPS) except for 3DMark 11, which will reflect an overall score, and a resolution of 1920 x 1080 will be used. All benchmarks will be tested with the most recent drivers, firmware and BIOS. Hi Tech Legion does not use a database to store its numbers for future use, so all the benchmarks you will see in this review have been performed within the time allotted.
Outline for testing:
- Benchmark the video card using the benchmark suite to see advantages or disadvantages in gaming performance compared to current video cards in the same price range.
- Performance Tune the video card using chosen performance tuning software
- Measure the heat output and subjectively compare the noise generated by the video card during idle and load.
- Resolutions Used:
- Single monitor - 2560 x 1600
- Surround - 5760 x 1080
- All temperatures taken with GPU-Z and will be read in Celsius.
- Noise Level:
- Noise Level will be evaluated using an electronic decimeter and read in decibels.
GeForce GTX Titan Overclocking video:
Fixed Function Benchmarks, sometimes called Synthetics benchmarks, are broken into two categories, Synthetic and Semi Synthetic.
- Synthetic: A Synthetic benchmark is created to test a specific feature and is developed using patterns and created files. These tests usually stress one section or module of the hardware to determine theoretical performance.
- Semi-Synthetic: A Semi-Synthetic benchmark is an application that has been created and used in trace form or specific patterns. These tests focus on an API, create bottlenecks on specific architectures then move to the next (CPU, GPU, Shaders, Tessellation, etc.). They can stress multiple modules at the same time to determine real game performance. Semi-Synthetic benchmarks require non specific drivers to verify (anti-cheat) and gather their scores based on traditional game engines.
HiTech Legion evaluates video cards using three, which are:
- 3DMark Fire Storm (Semi-Synthetic): Extreme, 2560x1600
- Unigine Heaven 3.0 (Synthetic): Ultra, Tessellation Extreme
- Unigine Valley (Synthetic): Ultra
Representing the forefront of gaming technology, DirectX 11 provides the stage for the latest and greatest PC hardware to flex their computing muscle. DirectX 11 titles are capable of unprecedented gaming realism with the use of features, such as Tessellation, which allow dynamic and scalable details to be applied in a scene by reducing or adding geometric complexity, or DirectCompute, which unlock the powerful parallel processing capability of GPUs. The latest 3D game engines can also utilize advanced anti-aliasing techniques, accurate depth of field applications, whole scene reflections, DirectCompute physics, and more. NVIDIA GPUs, starting with the "Fermi" line, and AMD GPUs, from the "Evergreen" series onward, are designed specifically to take advantage of these technologies.
- F1 2012: Ultra/MSAA=8X
- Sniper Elite V2: Ultra/AA=HIGH/AF=16/Adv.Shadows=OFF/ComputeShader=ON/Super Sampling=OFF
- Battlefield 3: Ultra Preset
- Metro 2033: Very High Preset/MSAA=4x/AF=16x/Tessellation=ON/DoF=ON/PhysX=OFF
- Call of Duty Black Ops II: Max Settings/16X CSAA/FXAA=ON
2D Surround (5760x1080)
Unlike previous generation GeForce GPUs, the GTX Kepler design is capable of driving up to four displays on a single GPU. A pair of dual-link DVI ports, an HDMI port and a DisplayPort can be mixed and matched requiring no extra dongles, unless you want to run 3D Vision surround, which requires a special dongle for the DisplayPort. The R300 drivers feature new desktop management software which enables Windows taskbar placement, bezel correction, bezel peak, fast center display acceleration and custom resolutions. As with previous GeForce cards, the NVIDIA GTX (Kepler) can be paired with one or two more GTX (Keplers), up to 3-way SLI for unparalleled gaming performance.
All settings for games benchmarked remain the same as previous single monitor benchmarks except for Metro 2033, where the Depth of Field was disabled.
Both Max Payne 3 and Crysis 3 require a demanding frame buffer in order to run at multiple monitor resolutions with any Anti-Aliasing. As you raise the Anti-Aliasing, more graphics memory is required in order to generate stable playable frame rates. Both the GTX 680 and GTX 690 (690 per gpu) have 2GB of GDDR5 256-bit interface, while the Titan has 6GB of GDDR5 384-bit interface. Memory Bandwidth for both the GTX 680 and 690 (690 per gpu) is 192 GB/s while the Titan has 288.4 GB/s. Both games Max Payne 3 and Crysis 3 were run at the same settings on all cards. Although not shown here Max Payne 3 @ 2X AA required 1662 MB of video memory (at below settings) and Crysis 3 1830 MB, with almost 400 MB of memory to spare Max Payne 3 ran smoothly on all cards but the GTX 680 being a single card actually showed some short spurts of lag as it played catch up on the frame buffer. The GTX 690 (dual GPU) had enough graphics power to conquer it's memory short comings. Crysis is where the obvious difference is shown, only leaving a little less than 200 MB of memory (680 and 690) the 680 was almost unplayable and the single GPU Titan virtually match the performance of the GTX 690. When Anti-Aliasing was raised over 2X the GTX 680 and 690 did not have enough memory bandwidth to run efficiently.
Provided by: NVIDIA
No Compensation was received for review of this product.
Class: Enthusiast Gamer
The NVIDIA GeForce GTX TITAN with an MSRP of $999.99 USD, falls into our Enthusiast Gamer Category.
Synthetics - NVIDIA's dual GPU GTX 690 has a commanding lead in all three benchmarks with the GTX TITAN in second place. While the GTX 690 has 697 points over the GTX TITAN's overall 3DMark Score, the gap is even greater between the GTX TITAN and the GTX 680, pulling ahead by 1386 overall points. The AMD HD7970 is only a hair's width ahead of the GTX 680 and both are considerably farther behind the GTX TITAN and the GTX 690.
Single Monitor Gaming Performance - Performance wise, the NVIDIA TITAN sits somewhere between the GTX 680 and the GTX 690, as expected, on average, although the gap between the 680 and the TITAN is much more pronounced compared to the gap between the GTX TITAN and the GTX 690. In F1 2012, the GTX TITAN actually pulls ahead of the GTX 690 with a 13 frame lead. Gameplay is noticeably smoother overall on the GTX TITAN, due to higher minimum framerates, especially at 2560x1600, where the larger frame buffer is an advantage.
Multi-Monitor Gaming Performance - The performance gap is even smaller when running a triple monitor 5760x1080 setup where a single GTX TITAN is only 15% shy, on average, compared to the dual-GPU GTX 690, while a single GTX 680 produced 48% less framerate. The HD 7970 does a bit better than the GTX 680 with only Metro 2033 dipping below unplayable frame rates but is still considerably dwarfed by the GTX TITAN performance.
Temperature, Power Consumption and Noise - A video card with this much power seems like it requires more than an 8-pin + 6-pin PCI-E connector but the GTX TITAN happily chugs along with a 250W TDP, matching a GTX 580 but outperforming it by a mile. Compared to the GTX 690, the cooling design has been revamped, utilizing a blower-style fan to draw cool air from the front and exhaust it all through the rear. The fan works in conjunction with a two-piece heatsink system comprised of a dense fin-array vapor chamber as well as a full-coverage heatsink which doubles as a cooling device for critical VRM components, as well as providing additional support for the PCB.
Features Check Off:
- Reference Board Design - Yes
- Reference Fan - Yes
- Aftermarket Board Design - No
- Aftermarket Cooling - No
- Aftermarket Power Control - No
- Factory OC - No
A reference based video cards is a video card whose specifications and features match that of what the OEM (in this case NVIDIA) has produced as a point of reference. This means that the board and PCB as well as cooling and power control match the exact specifications of the OEM. A non-reference board is designed and produced by an OEM partner with one common part and, in this case, the GPU. Non-reference cards may or may not have different power control but they will always have a non-OEM PCB and secondary architecture. Most non-reference boards are performance tuned by the OEM partner and come with aftermarket cooling. Some OEM partners choose to use a reference board design and cooling but factory tune the GPU and warranty that card to run at the increased frequency. HiTech Legion does not consider this to be a non-reference design. The same can be said for the OEM partners who may use strictly aftermarket cooling. HiTech Legion will benchmark OEM reference boards to establish a starting point of what can be expected in performance to allow the consumer the opportunity to evaluate what differences, if any, there may be when compared to OEM partner boards.
Forget what you know and have come to expect from video cards, as the GeForce GTX TITAN completely obliterates any doubts and goes well beyond anyone’s expectations on how a high-end video card should behave. We have seen, and have been impressed by, the efficiency of the Kepler architecture, as evidenced from the 600-series compared to 500-series GPUs, but the GeForce GTX TITAN pulls out even greater levels of efficiency that it almost seems like magic. NVIDIA has also sweetened the offering further by introducing GPU boost 2.0, showing that the green team not only has a solid grasp of what their Kepler architecture can do but also demonstrates how important software can be in further enhancing your hardware.
GPU Boost, which was initially introduced with the Kepler series launch, has been improved to provide greater gains by changing the monitoring algorithm to watch the GPU temperature target instead of GPU Boost 1.0’s GPU power target. This resulted in greater gains, especially for lower temperatures such as those achieved through liquid cooling and other enthusiast thermal management solutions. The GTX TITAN also has lower noise output than other 600-series cards, as the fan now has a tighter range of movement (as defined around the user target) when GPU Boost is engaged. As part of GPU Boost 2.0, users can now also “overclock” their monitors through NVIDIA control panel, allowing for pixel clock adjustments resulting in higher refresh rates. This is great news for users who play with VSync turned on, especially those with IPS or VA panels, so they can enjoy a smoother and more fluid gaming experience enjoyed by 3D-ready TN panel gamers without sacrificing the image quality advantage.
Power consumption is surprisingly low for such a powerful card, requiring only an 8-pin and a 6-pin PCI-E power connector, which is the same power requirement as a GeForce GTX 580 or a Radeon HD 7970. Combined with an efficient vapor chamber design and advanced fan control, the GTX TITAN is able to keep thermals in check while producing less noise than a reference GTX 680. Also, unlike the dual-GPU GTX 690, which blows exhaust on both front and back, the GTX TITAN has a rear external exhaust so it can be fitted in much smaller form factors. In fact, several system builders, such as iBUYPOWER, MAINGEAR, Falcon Northwest, Digital Storm and Origin PC, are offering custom slim and lower profile desktop builds featuring the GTX TITAN.
The GTX TITAN itself is not a large card, measuring only 10.5 inches in length, half an inch longer than a reference GTX 680 and half an inch shorter than an AMD Radeon HD 7970. Not only is it efficient in terms of power consumption and performance, but in PCB real estate as well, managing to fit the massive 551mm² GK110, a 6+2 phase digital PWM and 24 GDDR5 IC’s (12 on each side) for a massive 6GB frame buffer. It is essentially the same PCB as a K20X but tailored for desktop use. As an added bonus, the GeForce GTX TITAN retains full performance double-precision compute capability, easily toggled through the NVIDIA control panel. Compared to the GTX 680’s SMX with 8 double-precision CUDA cores, the GeForce GTX TITAN includes 64 double-precision CUDA cores per SMX, giving desktop PCs 1 teraflop of double-precision performance. This feature does not affect typical enthusiast gamers but makes GPU-accelerated development accessible to a larger number of users, especially those who do not have $3500 for a Tesla card.
With an MSRP of $999, the GTX TITAN is a big purchase decision, especially for a single video card, made complicated by the fact that a more powerful dual-GPU GTX 690 costs the same. A certain segment of the enthusiast market will undoubtedly find it hard to resist, however, considering performance can be pushed further beyond what two GTX 690s in Quad-SLI are capable of, via triple-SLI configuration. The cost is astronomical but the extra performance power is definitely within reach, if one can afford it. Once you also add in the fact that the GTX TITAN offers superior overclocking headroom, due to GPU Boost 2.0, offers more features, double-precision performance, larger frame buffer, consumes less power, produces less heat and runs quieter than a reference GTX 680, the choice becomes much clearer.
- GPU Boost 2.0
- 250W TDP (Only requires 1x 8-pin + 1x6-pin PCI-E connectors)
- GK110 GPU
- 6GB frame buffer
- Excellent thermal profile
- Produces less noise than a reference GTX 680
- Full double-precision functionality
- Only 10.5-inch long and fits most mid-tower cases
- Costs as much as a GTX 690