Kepler vs Tahiti – PC performance graphics benchmarks of Graphics Cards

Well, we’ve got a new single-chip flagship from NVIDIA in the form of the GeForce GTX 680. Since video cards are provided to us for constant testing, we are the final link in the summary test of graphics solutions, but we hope that our review will be very useful for our readers. Another good news is that our tests will now compare video cards from both manufacturers equally.

First look at GEFORCE GTX 680

NVIDIA, in its new product, the GTX680 video card, offers the latest architecture, announced back in 2010, named after the famous physicist Johannes Kepler. The new chip, produced using the latest 28 nm process technology, received the code name GK104. The use of new 28-nm process standards has made it possible to create a chip with a small chip area, lower power consumption and high performance compared to the previous generation.

 

According to NVIDIA, the GK104 is the fastest and most energy-efficient GPU.

Let’s look at the key features of the new GK104 chip:

• 28 nm production technology;
• Number of transistors 3.54 billion;
• Core area 294 mm2;
• Hardware support for DirectX 11, Shader Model 5.0, tessellation;
• Memory bus 256-bit;
• Standard core frequency 1006 MHz;
• Core frequency in TURBO mode – 1058 MHz;
• 8 streaming multiprocessors, 1536 scalar ALUs for floating point calculations;
• 128 texture addressing and filtering units with support for FP16 and FP32 components in textures and support for trilinear and anisotropic filtering for all texture formats;
• 4 ROP blocks, supporting antialiasing modes up to 32 samples per pixel, including FP16 or FP32 frame buffer format. Each block consists of an array of configurable ALUs and is responsible for generating and comparing Z, MSAA, blending;
• Integrated support for RAMDAC, two Dual Link DVI ports, as well as HDMI and DisplayPort.
• Integrated support for four monitors, including two Dual Link DVI ports, as well as HDMI 1.4a and DisplayPort 1.2
• PCI Express 3.0 bus support

The GeForce GTX 680 video card should replace its predecessor – the outdated model – GTX 580 based on the GF110 graphics processor.

Architecture and innovations in Kepler

The foundations of the Kepler architecture were laid back in Fermi (GeForce GTX 480), but the principle itself was completely redesigned due to increased demands for optimizing the efficient execution of complex computing tasks and very fast processing of geometry and tessellation.

Like Fermi, Kepler consists of several GPC (Graphics Processing Clusters) blocks, which are completely independent blocks within the video chip, operating as separate devices. The new GPU has 4 GPC blocks with two SMX streaming multiprocessors.

The main difference between SMX and the previous generation SM blocks is the increased number of shader blocks.

GPU Boost

One of the interesting features of the Kepler architecture is the now fashionable option of dynamically increasing/decreasing the operating frequency of the video chip. This technology, implemented by NVIDIA, is called GPU Boost. GPU Boost technology is a software and hardware complex that dynamically changes the frequency of the GPU based on its operating conditions.

 

The graphics chip constantly monitors power consumption, temperature, workload, and based on the data received, the operating frequency of the processor automatically changes, both upward and downward.

Full screen anti-aliasing using FXAA and TXAA methods

The firstborns of anti-aliasing – supersampling (SSAA) and multisampling (MSAA) – are being replaced by new anti-aliasing methods and algorithms – FXAA. The principle of this algorithm is the use of the computing power of stream processors. The new FXAA filtering method is up to 60% faster. compared to MSAA.

NVIDIA demonstrated the superiority of the FXAA method compared to MSAA at Kepler Editor’s Day in the well-known demo program – Samaritan, with support for all the capabilities of DirectX 11. Also, NVIDIA presented another anti-aliasing method – TXAA.

The main goal of TXAA anti-aliasing is the highest quality anti-aliasing, identical to the cinematic anti-aliasing used in 3D cartoons.

Adaptive VSync

Adaptive VSync is a software technology aimed at improving gaming comfort; the main task of the technology is to suppress image artifacts when the frame rate in the game becomes higher than the refresh rate of the monitor and artifacts appear in the form of frame gaps.

 

Adaptive VSync monitors frame rate and dynamically disables or enables VSync when the frame rate reaches the monitor’s refresh rate.

Hardware video encoding – NVENC

Chips with Kepler architecture have an NVENC hardware video encoding unit for H.264. The NVENC unit is capable of encoding FullHD video 4-8 times faster than real time and 4 times faster than CUDA-based methods.

Expanded support for multi-monitor output

Starting with the GeForce GTX 680, support for up to four image output devices is implemented simultaneously.

 

Also, the GTX680 supports 3D Vision Surround technology, outputting stereo images simultaneously on three monitors. To connect image output devices, the GK104 supports DisplayPort 1.2, HDMI 1.4a, high-definition “4K” monitors (3840×2160) and multi-stream audio output. 

Appearance

 

The length of the GeForce GTX 680 printed circuit board is 25.4 cm. On top of the card is covered with a cooling system casing, on the right side there is a centrifugal fan that pushes out hot air. The cooling system has also undergone some improvements – under the two-cell aluminum diffuser there are 3 copper heat pipes that effectively transfer heat to the diffuser, from which the heat flow is blown off by a fan. For effective heat dissipation, the radiator fins are located at an angle.

The fan motor incorporates sound-absorbing material, which effectively reduces noise.

Overclocking

For overclocking settings we used MSI Afterburner. As you can see from the screenshot below, we managed to achieve overclocking to a frequency of +150 MHz for the core and up to +500 MHz for the memory.

 

There was no point in driving it higher, since an increase of 10 MHz resulted in unstable operation of the graphics chip. Well, for those who are interested in maximum overclocking potential GeForce GTX 680, we have provided below a screenshot with a run in 3DMark11 with the maximum stable frequencies of the sample provided to us +160 MHz. The turbine speed was correspondingly turned up to maximum…

 

11000 parrots per GPU, simply amazing for a single-chip solution!

 Testing GEFORCE GTX680 

All video cards were tested at maximum graphics quality using MSI Afterburner. The purpose of the test is to determine how video cards from different manufacturers behave under the same conditions. The minimum and average frame size is taken as the performance indicator.

Our video cards were tested at separate screen sizes of 1920×1080 and 2560×1600 at the maximum graphics quality settings allowed by the games, and, of course, by the resources of the video cards. In this case, we took games that, at different quality settings, have the greatest impact on the performance drop of video cards.

Radeon HD 7970 tested at stock frequencies and overclocked Radeon HD 7970 OC – core 1160 MHz and memory 7160 MHz

Radeon HD 6970 tested at stock frequencies and overclocked Radeon HD 6970 OC – core 920 MHz and memory 5800 MHz

Radeon HD 6990 tested at stock frequencies and overclocked Radeon HD 6990 OC – core 880 MHz and memory 5500 MHz

GeForce GTX 580 tested at stock frequencies and overclocked GeForce GTX 580 OC – core 910 MHz and memory 4150 MHz

GeForce GTX 590 tested at stock frequencies and overclocked GeForce GTX 590 OC – core 680 MHz and memory 3800 MHz

GeForce GTX 680 tested at stock frequencies and overclocked GeForce GTX 680OC – core 1150 MHz and memory 7000 MHz

The test was carried out on one of the most powerful gaming configurations to date based on an Intel Core i7-3930K with overclocking to 4.8 GHz, eight gigabytes of DDR3 RAM with a frequency of 2133 MHz and an OCZ Vertex 3 SSD hard drive.

Metro 2033

Metro 2033 came out about two years ago and to this day is one of the most technologically advanced games, and, of course, one of the most demanding. This is the only game from 2010 that we took to test modern top-end video cards.

Testing at maximum quality settings 1920×1080

At the maximum graphics quality settings allowed by Metro 2033, the minimum that is needed is video cards Radeon HD 7970 or GeForce GTX 680. The Radeon HD 7970 is noticeably ahead in overclocking GeForce GTX 680.

Testing at maximum quality settings 2560×1600

At a resolution of 2560×1600, no video card can cope with Metro 2033.

Total War: Shogun 2

Total War: Shogun 2 is an excellent continuation of its Total War series that will appeal to many of its fans. Graphically, the game has become a leader in the abundance of graphical bells and whistles and special effects, representing a mixture of scale and beauty unprecedented in the world of strategy. The test was carried out with the maximum possible graphics settings in the game. Only the vertical sync function remained disabled. And just like in the 2011 video card test at a resolution of 2560×1600, video cards with a video memory capacity of at least two gigabytes were able to show themselves worthy.

Testing at maximum quality settings 1920×1080

GeForce GTX 680 showed very low FPS at these settings, losing even to its predecessor GeForce GTX 580, which turned out to be a very strange result for us, not consistent with other resources. Maybe we have some problems, but when monitoring the frequencies corresponded to the parameters and the loading of video cards was 99%…

Testing at maximum quality settings 2560×1600

With a resolution of 2560×1600, the only rational solution for this game is the Radeon HD 6990.

Crysis 2

Crysis 2 is also one of the most technologically advanced games to date. Of course, support for DirectX 11 effects was a little late, but this made it possible to slightly prolong the “live” interest in the game, thereby making it relevant for several months. The measurements, as in the 2011 game test, were carried out at the site of the bridge collapse.

Testing at maximum quality settings 1920×1080

At the maximum graphics quality settings allowed by Crysis 2 at a resolution of 1920×1080, the minimum and acceptable level of performance was shown by the Radeon HD 6970 or GeForce GTX 580. Single-chip GeForce GTX 6 is the fastest single video card in this game.

Testing at maximum quality settings 2560×1600

Again GeForce GTX 680 is the fastest single-chip solution…. 

The Witcher 2: The Assassins of Kings

The Witcher 2: Assassins of Kings has become one of the most beautiful, technologically advanced and spectacular games of last year. At the moment, the game is one of the most resource-intensive games of the present time. The game settings for the test were turned up to the maximum.

Testing at maximum quality settings 1920×1080

At a resolution of 1920×1080, with maximum graphics quality settings, the minimum level of performance will be shown by the Radeon HD 6970OC or GeForce GTX 580. At stock frequencies, the Radeon HD 7970 showed comparable performance to GeForce GTX 680, and in overclocking it overtook it.

Testing at maximum quality settings 2560×1600

At a resolution of 2560×1600, at the moment only overclocked dual-chip solutions can handle the game…

Battlefield 3

Battlefield 3 left a very pleasant gaming experience. Seamlessly combining the features of real combat and arcade mode. The test was conducted on ultra-high game settings. In graphic terms, the game is one of the most technologically advanced projects to date.

Testing at maximum quality settings 1920×1080

At the maximum graphics quality settings allowed by Battlefield 3 at a resolution of 1920×1080, the minimum level of performance was shown by the Radeon HD 6970 or GeForce GTX 580. The Radeon HD 7970 lost in stock GeForce GTX 680, and when overclocked again overtook it, however, only in terms of minimum FPS.

Testing at maximum quality settings 2560×1600

With a resolution of 2560×1600, the Radeon HD 7970 slightly strengthened its position, becoming the clear single-chip leader.

Batman: Arkham City

The graphics in Batman: Arkham City pleased with the implementation of DirectX 11, where tessellation is visible to the naked eye. And in place with PhysX, this makes the game simply unattainable in terms of graphics for consoles. The test was carried out at the maximum permissible settings for AMD cards without activating physics effects. The minimum FPS was not taken due to the imperfection of the benchmark.

Testing at maximum quality settings 1920×1080

At the maximum graphics quality settings allowed by Batman: Arkham City at a resolution of 1920×1080, all tested video cards coped with the benchmark without problems. The GeForce GTX 680 showed its architecture in all its glory, significantly surpassing even all previous dual-chip solutions.

Testing at maximum quality settings 2560×1600

With a resolution of 2560×1600, the GeForce GTX 680 further strengthened its position, outperforming the overclocked GeForce GTX 590 in stock.

Alan Wake

Well, we’ve reached the most resource-intensive games of 2012. The first of the candidates was the well-known Alan Wake. The game settings for the test were set to a high quality level.

Testing at maximum quality settings 1920×1080 

As we can see, at stock frequencies, the lead was GeForce GTX 680. When overclocked Radeon HD 7970 showed the most significant performance among single-chip solutions.

Testing at maximum quality settings 2560×1600

At a resolution of 2560×1600 we see a similar picture. In stock ahead GeForce GTX 680, and overclocked – Radeon HD 7970.

Gas Guzzlers Beta

 

At the moment, in our opinion, the Gas Guzzlers demo is one of the most advanced racing series in terms of graphics. and although the release of the full version is still very far away, the demo version looks very impressive. The game settings for the test were turned up to the maximum.

Testing at maximum quality settings 1920×1080 

In this game we see the complete superiority of Tahiti architecture video cards, which show simply impressive performance when overclocked.

Testing at maximum quality settings 2560×1600

At a resolution of 2560×1600, at the moment only an overclocked version of the Radeon HD 7970 can handle the game…

King Arthur II – Demo

King Arthur II is currently an excellent indicator of negligence in terms of game optimization. Of course, we understand that the game is a PC exclusive, but not to the same extent…

Testing at maximum quality settings 1920×1080 

And in this game all the delights of the new generation of video cards are revealed. And as always, in nominal terms it’s faster GeForce GTX 680, and overclocked – Radeon HD 7970.

Testing at maximum quality settings 2560×1600

At a resolution of 2560×1600, GeForce GTX 680 i Radeon HD 7970 show absolutely identical results.

Risen 2 Dark Waters

Risen 2 Dark Waters is the final link in today’s game test. The game came out not so long ago and at high screen levels it has a significant impact on video cards.

Testing at maximum quality settings 1920×1080 

As we see, the new generation of video cards is almost on par. GeForce GTX 680 at this resolution has a slight advantage over its rival.

Testing at maximum quality settings 2560×1600

At a resolution of 2560×1600 the picture is almost identical, except when overclocked  The Radeon HD 7970 turns out to be slightly faster.

3DMark 11

3DMark 11 is the most optimal solution for testing modern top-end video cards, since its performance is not limited by the central processor. The result was the result of GPU speed in modes extreme and performance.

Testing at quality settings 1280×720 Performance

Testing at quality settings 1920×1080 Extreme

In all cases, the GeForce GTX 680 showed simply impressive performance in 3DMark 11, comparable to modern dual-chip solutions.

So, in the end, we did not decide to clutter our review with FPS calculations and the conclusion of various price advantages. At the moment, the GeForce GTX 680 and Radeon HD 7970 occupy almost the same price niche and you can easily buy both products at a comparable price.

As for performance, in some games the GeForce GTX 680 is faster, and in others the Radeon HD 7970, in others they showed comparable results. The overclocking level of both cards is quite high, so overclocking enthusiasts will be happy with their purchase.

If we talk about the advantages, then each solution has its advantages. Thus, the Radeon HD 7970 gives the user a slightly higher overclock relative to the base clock frequency. But owners of the GeForce GTX 680 will receive a more energy-efficient single-chip solution. Both video cards are worthy solutions and share the first place podium between two.

Well, the most interesting thing awaits us yet. Users buy new video cards not to replay games of bygone days, but with the prospect of the near future, when with the release of a new project they will be able to get maximum pleasure from the gameplay. So stay with us and you will always be aware of the performance of modern graphics solutions exactly when you really need it!