“... It is impossible to call the GTX 680 the undisputed leader. In terms of performance, the card is not much ahead of the HD 7970. Perhaps the only advantage of Kepler is manufacturability. Automatic overclocking by stock TDP, original smoothing modes, the ability to play
Gambling https://www.site/ https://www.site/
NVIDIA has entered the next generation graphics card race. At the end of March this year, the GeForce GTX 680, the main competitor of the AMD Radeon HD 7970, went on sale. Now we will tell you what is hidden under the code name Kepler, how many cores a fresh crystal received and at what frequency it operates. And, of course, we will answer the age-old question: who is faster, "green" or "red"?
Reorganization
Architecture Fermi, which has been betting on for the last couple of years, has exhausted itself. It was replaced by Kepler. At first glance, everything is the same: stream processors are combined into SM modules, which, in turn, form GPC clusters, and a GPU is assembled from them. But if in GF110 (GeForce GTX 580) each SM-module was equipped with 32 CUDA-cores, then in Kepler there were six times more of them - 192 pieces. The thicker blocks (now called SMX) and auxiliary elements have also been added: there are now 16 “texture units” instead of 4, and sets of special operations - 32 against 4 SFU from the previous flagship.
A pair of SMXs and one rasterizer form a GPC (Graphics Processing Cluster). Four such sets form the basis of the chip GK104- the hearts of the GTX 680. Thus, the new GPU carries 1536 cores and 128 texture units, while the GF110 was content with 512 processors and 64 TMUs. But the number of PolyMorph Engines responsible for tessellation has decreased: 8 versus 16 in the previous generation. Few? No matter how. The developers not only revised the principles of operation of the PM Engine, but also added megahertz. So, according to the engineers, the overall performance increased by about 28%.
But what Kepler abandoned is frequency separation. At Fermi, the stream processors ran at twice the speed of the stone, now the numbers are even. On the one hand, this is good: the main elements of GK104 received a hefty dose of adrenaline. On the other hand, the total power of a single CUDA core has noticeably dropped.
Another loss is the memory bus. GK104 is equipped with four 64-bit GDDR5 controllers, so the bit depth is 256 versus 384 bits for GF110 and Thaiti on the Radeon HD 7970. Now the ROP is also smaller: 32 instead of 46 units for the GTX 580.
New opportunities
With Kepler, several new technologies have been introduced. First - GPU Boost. It monitors temperature and power consumption and, if the indicators do not exceed critical values, automatically increases the speed and voltage of the crystal. All this works at the hardware level and does not turn off even during manual overclocking.
The second is the ability to connect four monitors at once. At the same time, for 3D Vision Surround(games on three displays) no more need for SLI-bundle, one GTX 680 is enough.
The third is a discrete video encoding chip NVENC. NVIDIA cards have been good at processing video streams before, but they used the processing power of CUDA cores for this. Now they play only an auxiliary role: the new module performs the main calculations and does it four times faster than Fermi. The weak point of NVENC is that it is only familiar with the H.264 codec.
Fourth - fresh anti-aliasing algorithms, TXAA and FXAA. The latter is a "cheap" replacement for the traditional MSAA. The impact on performance is minimal, but you have to pay with a slight blurring of the picture. TXAA is a more serious approach, and according to the developers, it not only surpasses MSAA in quality, but also requires less resources.
Finally, the fifth is the adaptive vertical sync mode. Everything is simple here: VSync is activated only when fps exceeds the screen resolution. As a result, subsidence of the frame rate below the coveted threshold no longer leads to a sharp drop in speed.
cold blooded
Despite a threefold increase in the number of stream processors, the number of transistors used to build a chip increased slightly - from 3 to 3.54 billion. Thanks to the new 28-nm process technology, the core area and power consumption turned out to be small. Even at maximum load, TDP does not exceed 195 watts, and this is when the base frequency reached a fantastic mark of 1006 MHz, and under GPU Boost it rises to 1058 MHz! For comparison, a Radeon HD 7970 running at 925 MHz and carrying 4.31 billion transistors consumes 250 watts.
The cool nature of the GK104 made it possible to simplify the design of the board. The GTX 680 does not have an 8-pin PCIe connector, so loved by top solutions - a pair of 6-pin sockets was enough. Only four phases were allocated to power the crystal, and, judging by the wiring, the fifth was abandoned at the last moment. The length of the card is 25 cm, which is slightly less than the flagship GeForce of previous generations.
The novelty is equipped with eight GDDR5 modules with a total capacity of 2 GB. The memory speed is a record 6008 MHz, which partially compensates for the low bus width. Back panel equipped with one HDMI 1.4a with DisplayPort 1.2 and a pair of DVI. The interface is PCIe 3.0, backwards compatible with PCIe 2.0.
A two-story turbine is responsible for cooling. It includes a large aluminum radiator with three heat pipes and an unusual for "green" cylindrical turntable, which uses sound-absorbing materials.
First in the world
We received the payment directly from the Russian office of NVIDIA. Appearance standard cards: the entire surface is covered with a black plastic casing and decorated with company logos. In terms of dimensions, the GTX 680 is really a little smaller than its predecessor and looks very neat. Of non-standard solutions, we note the unusual arrangement of power contacts: two 6-pin sockets are not in line, but one above the other.
For tests, we assembled a stand based on the motherboard Gigabyte GA-X58A-UD3R. Used as a processor Core i7-920, memory set three bars Kingston HyperX DDR3-1666MHz 2 GB each Windows 7 Ultimate 64-bit and all programs recorded on Kingston SSDNow.
The list of applications includes 3DMark11, Unigine Heaven Benchmark 2.5, just cause 2 , DiRT 2, Aliens vs. Predator, Batman Arkham City and Total War: Shogun 2. The competitors were appointed GeForce GTX 590, GTX 580, AMD Radeon HD 6970 and HD 7970.
Who is first?
In 3DMark11, the new product was 17% faster than the HD 7970 and came close to the results of the dual GTX 590! Kepler was also liked by the resource-hungry Unigine Heaven Benchmark 2.5: superiority over AMD - 26%. In games, the numbers are even more interesting.
Aliens vs. The Greens' flagship Predator hit 57.7 fps, beating the GTX 580 with the HD 6970 and trailing the HD 7970 and GTX 590 by 1.5 and 15.8 frames, respectively. The newcomer also lost in Just Cause 2, which is optimized for AMD drivers, - the difference with the HD 7970 was 21.7 fps. We managed to distinguish ourselves in DiRT 2, the "red" leader missed 37.5 fps.
NVIDIA's next win is Batman: Arkham City. With PhysX, MSAA 8x and a resolution of 1920x1080 turned on to the maximum, the hero of our review gave out quite playable 26 fps, while the HD 7970 barely reached 20 frames. After disabling PhysX, the situation did not change: with MSAA 4x, the gap was 4-6 fps, and after switching to 8x, it was 15-18 fps. The triumphant march of the GTX 680 was stopped by Total War: Shogun 2. The older GeForce reached 23.2 fps, the HD 7970 - up to 24.9.
The last word
Let's face it, we expected more from the GeForce GTX 680. Look at the resulting fps ratio: only 8% faster than the HD 7970. The same difference was between the GTX 580 and HD 6970 two years ago. Yes, in many applications the "green" flagship is head and shoulders above its competitor. But all these are just NVIDIA-optimized games. In impartial tests - Aliens vs. Predator, Batman: Arkham City, Total War: Shogun 2 - rivals are on par.
Today's victory for the GTX 680 is quite possibly due to the late release. If the GeForce appeared earlier than the HD 7970, AMD engineers would bleed, but they would have pulled the Graphics Core Next up to Kepler frequencies. What would come of it - see our plates. We overclocked the HD 7970 to the average GK104 speed of 1035 MHz and got similar results to the GTX 680.
It is impossible to call the GTX 680 the undisputed leader. In terms of performance, the card is not much ahead of the HD 7970. Perhaps the only advantage of Kepler is manufacturability. Automatic overclocking by TDP stock, original anti-aliasing modes, the ability to play on three monitors, a separate video decoding chip, PhysX, 3D Vision, noise isolation of the cooling system. And all this with an extremely modest power consumption of 195 watts. Otherwise, AMD and NVIDIA are roughly equal. And now it all depends on competent support, driver settings and, of course, the pricing policy of companies. For example, HD 7970 can already be bought for 16,500 rubles, official the price tag for the GTX 680 is 17,990 rubles.
Thinking out loud
If we digress a little from the results of Kepler and take a closer look at the characteristics, we can see a lot of interesting things. The GTX 680 looks too weird for a high-end graphics card. The first thing that catches your eye is a noticeable simplification of the memory subsystem: only 256 bits versus 384 bits for the GTX 580. More to come. The power consumption of the novelty is surprisingly small - 195 watts. And while this is in line with the current trend of increasing performance per watt, we all know that NVIDIA has never been shy about making big, hot chips that consume 250W. The suspiciously modest amount of GDDR5 is also confusing - only 2 GB, and this is for a board that should provide a crazy resolution of 5760x1080 when playing on three monitors. By the way, about the latter: GK104 supports only four displays at the same time, although it could well pull six, like the older representatives of AMD. And, finally, a negligibly small increase in elements and a modest core area. And after all NVIDIA never aspired to minimalism.
Drawing parallels with past generations, one gets the feeling that we have a stripped-down version of the crystal and soon we may well expect something more powerful. Of course, these are just thoughts out loud, and perhaps NVIDIA really went against their own principles, but the GTX 680 is very similar to some kind of GTX 670.
In the main part of the article, we mentioned several new technologies. Let's talk about them in more detail. Let's start with Adaptive VSync.
Every monitor has natural frequency sweep. For modern LCD panels it is 60 or 120 Hz with 3D Stereo support. These numbers indicate the maximum number of frames that the screen can show in one second. However, in some games, fps exceeds this limit. On the one hand, it's good - no brakes. On the other hand, overscan results in image twitching.
VSync allows you to adjust the refresh rate to the monitor's capabilities. If the matrix is limited to 60 Hz, then the video card will not give out more than 60 fps, which means there will be no image distortion. But this approach has one significant disadvantage. When the GPU cannot provide the required number of frames, VSync sets the bar "no more" than 30, then 20 and 10 fps. And he does not pay attention to the fact that the card easily holds, for example, 40 fps. NVIDIA has solved this problem.
Adaptive VSync monitors system performance, and if the speed drops below 60 fps, it disables VSync, preventing the board from being idle. As soon as the game gains momentum, the restriction returns. Bottom line: no artifacts and artificial freezes.
Smoothing
Next on the agenda are new antialiasing methods designed to replace the usual Multisampling Antialiasing. To understand why the standard MSAA is bad, you need to understand the principles of its work.
The technology was invented to get rid of aliasing at the corners of objects. Unpleasant notches appear for a very simple reason. The LCD panel can be imagined as a sheet of a notebook in a small cell, each of which can be painted over in one color. Try to draw some kind of house according to these rules: horizontal and vertical lines will turn out to be even, and slanted lines will be herringbone. Everything is the same in the computer. The video card turns the shapes into pixels (ROP blocks do this) and fills the screen with them. Since there are few dots on the display, characteristic “teeth” come out on uneven planes. Getting rid of them is relatively simple: you need to recolor the pixels adjacent to the edges of objects in transitional colors. To determine them - and use anti-aliasing.
At first, SuperSampling was used for this: the scene was rendered at a resolution 2/4/8 times higher than necessary, then the picture was compressed to the capabilities of the matrix and displayed on the monitor - the notches disappeared. One thing is bad - this approach consumed an indecently many resources, so Multisampling soon replaced it. With the help of cunning algorithms, he processed only problem areas and did not touch the rest of the image, which made it possible to significantly increase performance.
Unfortunately, with the increase in the number of polygons and the emergence of new complex effects, this technology turned out to be ineffective: the loss in speed no longer corresponded to the increase in quality. Plus, many problems began to create engines with deferred rendering. They first calculate the geometry and only then apply the lighting. A vivid example is Metro 2033, where the inclusion of MSAA led to a monstrous drop in fps. In general, there are enough difficulties with modern anti-aliasing, and manufacturers are looking for an alternative to it.
The first sign was Morphological AA, announced with AMD Radeon HD 6970. The video card renders a frame in standard resolution and applies a light "photoshop" to it - it blurs the image. As a result, there are no "jaggies", but the clarity of the textures is lost. The same technique is now used by NVIDIA, although it is called FXAA.
But TXAA is already something new. It combines both traditional anti-aliasing methods and post-processing based on HDR and information from previous frames. The final quality result exceeds MSAA 8x, and consumes less resources than MSAA 4x.
Note that all these innovations are not games of marketers. We did a test comparing MSAA and FXAA. The performance difference was 10%. You can see the result in the screenshots below. FXAA does not look as beautiful as traditional MSAA, but in dynamics it is almost not noticeable.
Unfortunately, we were unable to test the brand new TXAA. It must be built into the game; its support in future projects has already been announced epic, Crytek, gear box and CCP.
Overclocking
GPU Boost, implemented in the GTX 680, has long been requested for video cards. In central processors, automatic overclocking has been used since the days of the first Core i7. It entered the GPU only with the advent of the GTX 580, and even then in a truncated form. The technology was intended to protect against overheating: it lowered the speed when TDP was exceeded. This is how NVIDIA protected its chips from special heaters like furmark and could significantly increase the final frequency of work. The downside of the old approach was the implementation: everything was controlled by drivers that responded to pre-defined applications.
This bug has been fixed by AMD. PowerTune, introduced with the Radeon HD 6970, also struggled with overheating, but the temperature was monitored by on-chip sensors, which turned out to be much quicker. software solutions and independent of specific programs.
With the GTX 680, the engineers went even further. Hardware controls now overclock the GPU. When the game cannot fully load the stone and there is a TDP headroom, the performance of the board increases. As soon as the maximum level of power consumption is reached, the card slows down. Interestingly, the declared 1056 MHz is not the limit for GPU Boost. In our tests, Kepler often rose to 1100 MHz.
The only negative point is that “doping” cannot be turned off. Even with an increase in the base frequency, GPU Boost continues to work. So, setting the Base Clock to 1100 MHz, we witnessed how the GK104 got to 1300 MHz! True, only professional testers will complain about this, a “free” increase will never hurt us.
Physics
Another piece of news is the PhysX update. With the GTX 680, he got two fresh effects. The first is a modified version of the hair calculation. If individual strands were shown on the GTX 580, then at the Kepler presentation we were shown a technical demo with a natural yeti, each hair of which reacted to external influences like wind or stroking.
The second is the possibility of destruction of objects. NVIDIA taught the engine to break objects in real time. And not just calculate the animation of pre-prepared blocks, but naturally break entire columns, apply textures to the resulting fragments and scatter them according to all the laws of physics.
Table 1.
Specifications
|
NVIDIA GeForce GTX 690 |
|
|
AMD Radeon HD 7970 |
|
|
AMD Radeon HD 7950 |
|
|
AMD Radeon HD 6970 |
|
|
NVIDIA GeForce GTX 580 |
|
|
NVIDIA GeForce GTX 590 |
|
|
Number of transistors |
|
|
NVIDIA GeForce GTX 690 |
|
|
AMD Radeon HD 7970 |
|
|
AMD Radeon HD 7950 |
|
|
AMD Radeon HD 6970 |
|
|
NVIDIA GeForce GTX 580 |
|
|
NVIDIA GeForce GTX 590 |
|
|
Process technology |
|
|
NVIDIA GeForce GTX 690 |
|
|
AMD Radeon HD 7970 |
|
|
AMD Radeon HD 7950 |
|
|
AMD Radeon HD 6970 |
|
|
NVIDIA GeForce GTX 580 |
|
|
NVIDIA GeForce GTX 590 |
|
|
Number of stream processors |
|
|
NVIDIA GeForce GTX 690 |
|
|
AMD Radeon HD 7970 |
|
|
AMD Radeon HD 7950 |
|
|
AMD Radeon HD 6970 |
|
|
NVIDIA GeForce GTX 580 |
|
|
NVIDIA GeForce GTX 590 |
|
|
Graphics core frequency |
|
|
NVIDIA GeForce GTX 690 |
|
|
AMD Radeon HD 7970 |
|
|
AMD Radeon HD 7950 |
|
|
AMD Radeon HD 6970 |
|
|
NVIDIA GeForce GTX 580 |
|
|
NVIDIA GeForce GTX 590 |
|
|
Stream processor frequency |
|
|
NVIDIA GeForce GTX 690 |
|
|
AMD Radeon HD 7970 |
|
|
AMD Radeon HD 7950 |
|
|
AMD Radeon HD 6970 |
|
|
NVIDIA GeForce GTX 580 |
|
|
NVIDIA GeForce GTX 590 |
|
|
Type, memory size |
|
|
NVIDIA GeForce GTX 690 |
|
|
AMD Radeon HD 7970 |
|
|
AMD Radeon HD 7950 |
|
|
AMD Radeon HD 6970 |
|
|
NVIDIA GeForce GTX 580 |
|
|
NVIDIA GeForce GTX 590 |
|
|
Memory frequency |
|
|
NVIDIA GeForce GTX 690 |
|
|
AMD Radeon HD 7970 |
|
|
AMD Radeon HD 7950 |
|
|
AMD Radeon HD 6970 |
|
|
NVIDIA GeForce GTX 580 |
|
|
NVIDIA GeForce GTX 590 |
|
|
Data bus |
|
|
NVIDIA GeForce GTX 690 |
|
|
AMD Radeon HD 7970 |
|
|
AMD Radeon HD 7950 |
|
|
AMD Radeon HD 6970 |
|
|
NVIDIA GeForce GTX 580 |
|
|
NVIDIA GeForce GTX 590 |
|
|
Number of texture blocks |
|
|
NVIDIA GeForce GTX 690 |
|
|
AMD Radeon HD 7970 |
|
|
AMD Radeon HD 7950 |
|
|
AMD Radeon HD 6970 |
|
|
NVIDIA GeForce GTX 580 |
|
|
NVIDIA GeForce GTX 590 |
|
|
Number of rasterization blocks |
|
|
NVIDIA GeForce GTX 690 |
|
|
AMD Radeon HD 7970 |
|
|
AMD Radeon HD 7950 |
|
|
AMD Radeon HD 6970 |
|
|
NVIDIA GeForce GTX 580 |
|
|
NVIDIA GeForce GTX 590 |
|
|
power usage |
|
|
NVIDIA GeForce GTX 690 |
|
|
AMD Radeon HD 7970 |
|
|
AMD Radeon HD 7950 |
|
|
AMD Radeon HD 6970 |
|
|
NVIDIA GeForce GTX 580 |
|
|
NVIDIA GeForce GTX 590 |
|
|
Board length |
|
|
NVIDIA GeForce GTX 690 |
|
|
AMD Radeon HD 7970 |
|
|
AMD Radeon HD 7950 |
|
|
AMD Radeon HD 6970 |
|
|
NVIDIA GeForce GTX 580 |
|
|
NVIDIA GeForce GTX 590 |
|
|
Interface |
|
|
NVIDIA GeForce GTX 690 |
|
|
AMD Radeon HD 7970 |
|
|
AMD Radeon HD 7950 |
|
|
AMD Radeon HD 6970 |
|
|
NVIDIA GeForce GTX 580 |
|
|
NVIDIA GeForce GTX 590 |
|
|
Price for January 2012 |
|
|
NVIDIA GeForce GTX 690 |
17 990 rubles |
|
AMD Radeon HD 7970 |
16 500 rubles |
|
AMD Radeon HD 7950 |
15 000 rubles |
|
AMD Radeon HD 6970 |
10 000 rubles |
|
NVIDIA GeForce GTX 580 |
12 500 rubles |
|
NVIDIA GeForce GTX 590 |
23 500 rubles |
Table 2.
Synthetic tests
3DMark11
|
NVIDIA GeForce GTX 680 |
|
|
VTX Radeon HD 6970 |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
Unigine Heaven Benchmark 2.5
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
Table 3
Gaming benchmarks (fps)
Aliens vs. Predator (DX11)
|
Very high. 1680x1050, AF 16x, AA 2x |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
Very High, 1920x1080, AF 16x, AA 2x |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
DiRT 2 (DX11)
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
Just Cause 2
|
High, 1680x1050, AF 16x, AA 4x |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
High, 1920x1080, AF 16x, AA 4x |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
Batman: Arkham City (DX11, Full PhysX)
|
High, 1680x1050, AF 16x, AA 4x |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
High, 1680x1050, AF 16x, AA 8x |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
High, 1920x1080, AF 16x, AA 4x |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
High, 1920x1080, AF 16x, AA 8x |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
XFX R7970 Double Dissipation (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
Batman: Arkham City (DX11, no PhysX)
|
Ultra. 1680x1050, AF 16x, AA 4x |
|
|
NVIDIA GeForce GTX 680 |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
Ultra, 1680x1050, AF 16x, AA 8x |
|
|
NVIDIA GeForce GTX 680 |
|
|
ASUS HD 7970 DirectCU II Top (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
Ultra, 1920x1080, AF 16x, AA 4x |
|
|
NVIDIA GeForce GTX 680 |
|
|
ASUS HD 7970 DirectCU II Top (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
Ultra, 1920x1080, AF 16x, AA 8x |
|
|
NVIDIA GeForce GTX 680 |
|
|
ASUS HD 7970 DirectCU II Top (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
ASUS HD 7970 DirectCU II Top (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
Total War: Shogun 2 (DX11)
|
Ultra. 1680x1050, AF 16x, AA 4x |
|
|
NVIDIA GeForce GTX 680 |
|
|
ASUS HD 7970 DirectCU II Top (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
Ultra, 1680x1050, AF 16x, AA 8x |
|
|
NVIDIA GeForce GTX 680 |
|
|
ASUS HD 7970 DirectCU II Top (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
Ultra, 1920x1080, AF 16x, AA 4x |
|
|
NVIDIA GeForce GTX 680 |
|
|
ASUS HD 7970 DirectCU II Top (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
Ultra, 1920x1080, AF 16x, AA 8x |
|
|
NVIDIA GeForce GTX 680 |
|
|
ASUS HD 7970 DirectCU II Top (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
ASUS HD 7970 DirectCU II Top (925/5500 MHz) |
|
|
ASUS HD 7970 DirectCU II Top (1035/5500 MHz) |
|
|
VTX Radeon HD 6970 |
|
|
ZOTAC GeForce GTX 580 AMP! Edition (772/4008 MHz) |
|
|
Point of View GeForce GTX 590 |
|
Table 4
Price/performance ratio
|
Performance |
|
|
NVIDIA GeForce GTX 680 |
|
|
AMD Radeon HD 6970 |
|
|
NVIDIA GeForce GTX 580 |
|
|
NVIDIA GeForce GTX 590 |
|
|
NVIDIA GeForce GTX 680 |
|
|
AMD Radeon HD 7970 (925/5500 MHz) |
|
|
AMD Radeon HD 7970 (1035/5500 MHz) |
|
|
AMD Radeon HD 6970 |
|
|
NVIDIA GeForce GTX 580 |
|
|
NVIDIA GeForce GTX 590 |
|
Due to the built-in frequency control algorithm GPU and voltage, the GTX 680 is overclocked in a slightly different format. No graphs and measurements can be made, due to the GPU floating frequency and voltage. First, let's figure out what the GTX 680 overclocking tools are. It's worth noting here that the usual methods for checking stability - Heaven, Furmark and others, no longer work. The overclocker's arsenal now includes the ability to control the maximum power consumption, the standard wording of which is 100%, or ~195 watts for the whole card!
From now on, the frequency of 1006 MHz is the base frequency, below which the graphics core does not fall under any load. Be it Furmark or any other extreme test. The "Turbo" (GPU Boost) value is 1058 MHz, indicating the average GPU frequency over the results of most gaming scenes. The function of the same name, focusing on the maximum allowable load and the temperature reached, automatically adjusts the frequency. As visual aid consider three simple examples how this technology works in 3D applications.
advertising
After starting, GPU Boost sets the frequency to 1110 MHz, the voltage is 1.175 V, as the video card heats up, the optimism of the algorithm decreases, and consumption increases at the same time. The final value is 1084 MHz at 1.137 V.
The frequency and voltage at the beginning of the test rise to the same 1110 MHz and 1.175 V, after a while decreasing to 1006 MHz and 1.075 V. Hard Reset turned out to be almost the only game where the Turbo algorithm considers the load provided too large, reducing the parameters to the above values. In the rest gaming applications the GPU frequency was almost always in the range from 1084 to 1120 MHz.
advertising
Furmark
Is it clearly visible how the power consumption protection works, and the frequency does not fall below 1006 MHz, or is it just hidden? The range of power consumption jumps from 117% at 1.087 V to 125% at 1.075 V. In fact, the video card reaches its programmed maximum and triggers protection functions that lower the frequency or voltage, or other yet unknown methods are used.
If you do not take any actions with overclocking, but raise the maximum power consumption threshold to 132%, which is approximately equal to 260 watts, then the following GPU Boost behavior is observed.
Unigine Heaven Benchmark v2.5
After starting, out of habit, a frequency of 1110 MHz and a voltage of 1.175 V are set, as we test, the GPU frequency value, taking into account the extended power consumption limit, becomes 1097 MHz (at 100% it was 1084 MHz), the voltage is 1.162 V (1.137 V.) It is easy to understand that " Turbo" is also capped at the top. And overclocking, albeit with an increased limit, can hardly be called significant - 1097 MHz versus 1084 MHz.
hard reset
132% "doping" allowed GPU Boost to assign a frequency of 1097 MHz at 1.162 V for most of the test (at 100% it was 1006 MHz and 1.075 V), but at times the values dropped to 1058 MHz and 1.125V. In general, there is a sense from the increased limit, especially in those cases where the default frequency increases slightly. Hard Reset is one of those games.
The upper protection bar has expanded from 125% to 135%, but this did not affect the overall level of power consumption, only 15 watts to the nominal frequencies and settings. During the test, the GPU frequency floated between 1033 MHz and 1058 MHz several times, but most of the time it was 1045 MHz. Together with the frequency "Turbo" operates with voltage.
Behind the scenes is the Kepler temperature. In theory, the maximum is 82-83°C, the cooling fan constantly adjusts to changing test conditions, not allowing the GPU to warm up above the given numbers. Naturally, GPU Boost also takes temperature into account. And if the video card, for some reason, exceeds a certain figure, then protection technologies will come into play. The practical maximum, according to NVIDIA, is 98°C - this is the critical temperature after which the system can completely turn off, having previously used all available protection methods.
advertising
In the bowels of the program hides the most important tool for an overclocker - access to voltage regulation.
The maximum allowable vGPU voltage is 1.150 V, and depending on the quality of components, heating, VRM / PWM calibrations, it can fluctuate +/-5%. On this instance of the video card, the measurement showed 1.123 V.
But keep in mind that no one has canceled the upper limit of 132% of the nominal! As a result, the ideal overclocking option for Kepler looks like this. Take a game that loads the video card well and run it, focusing on the indicators of the diagnostic program. Gradually increase the frequency of the GPU without touching the voltage. Alas, "Turbo" cannot be turned off, it will try to adjust the frequency as long as there is a margin for power consumption. Finding the right balance point is difficult, but possible. And then in most graphics applications the accelerator will stabilize at your desired frequency.
In the end, I managed to find the frequency at which the "Turbo" came into action, but not much, only +10 MHz. The voltage was at 1.150 V (actually 1.23 V), the power consumption threshold was +132%, the GPU frequency was +170 MHz, the fan speed was set so that the temperature did not rise above 70°C. In all test scenes, I had to monitor the frequency every second, but the result exceeded expectations - a stable 1280 MHz.
advertising
Even before the start of testing, I was sure that the Turbo technology would bring a lot of trouble to enthusiasts, but as it turned out, in vain. It is impossible to bypass it, so it is better to make friends with it. For ordinary users, it will be very useful, for example, set 132% and forget, the algorithm itself will select the upper limit depending on the type of load on the video card.But in any case, we are talking about frequencies of the order of 1.1 GHz. And enthusiasts who do not mind the time and patience will definitely be rewarded, they will be able to get closer to 1.3 GHz. Don't forget about memory. Its voltage is much higher than that used by a competing video card. On the AMD Radeon HD 7970, the Hynix chips get 1.53-1.55 V, while on the NVIDIA GTX 680 it's all 1.65 V, which resulted in an impressive overclock of 1800 MHz (7200 MHz). Given the 256-bit bus, I expect a decent performance boost in quality test modes.
Tools and methodology for performance tests
In some games, where possible, built-in performance measurement tools were used:
- 3Dmark 2011 - Extreme 2560x1440, 1920x1080;
- Unigine Heaven Benchmark v2.5;
- Colin McRae DIRT II;
- Total War Shogun II;
- Colin McRae Dirt III
- Batman: Arkham City
- F1 2011;
- hard reset.
For the following games, performance was measured using the FRAPS v3.4.7 utility:
- Metro 2033;
- Aliens vs Predator 3;
- StarCraft II;
- Battlefield III;
- Crysis 2;
- TESV Skyrim;
- The Witcher 2;
- Deus Ex - Human Revolution.
advertising
VSync was disabled during testing.To avoid errors in the measurement error, all tests were performed three times. When calculating the average FPS, the arithmetic mean of the results of all passes was taken as the final result.
Test results
List of video cards and their frequencies:
| Name | Rated frequencies GPU/Mem, MHz | Overclocking GPU/Mem, MHz |
| GTX 680 | 1000/1500 | 1280/1800 |
| GTX 590 | 607/854 | 650/900 |
| GTX 580 | 772/1002 | 950/1100 |
| GTX 570 | 732/950 | 900/1050 |
| HD 7970 | 925/1375 | 1250/1700 |
| HD 7950 | 800/1250 | 1225/1800 |
| HD 7770 | 1000/1125 | 1200/1300 |
| HD 7750 | 800/1125 | - |
| HD 6990* | 880/1250 | - |
| HD 6990 | 830/1250 | 950/1450 |
| HD 6970 | 880/1375 | 1000/1500 |
| HD 6950 | 800/1250 | 1000/1500 |
| HD 6870 | 900/1050 | 1000/1200 |
| HD 6850 | 775/1000 | 1000/1150 |
advertising
The performance of the NVIDIA GTX 680 is taken as the starting point of reference.3D Mark 2011
Settings:
- Extreme profile;
- DirectX11;
- Resolution - 1920x1080; 2560x1440;
- Four graphic tests;
- Results: GPU Score.
GPU Score
2560x1440 | 1920x1080GPU Score
Overclocking2560x1440 | 1920x1080
Please enable JavaScript to see charts
advertising
| Model | Medium 1920% | Medium 2560% | Medium |
| GTX 680 | 0 | 0 | 0 |
| GTX 590 | 3 | 7 | 5 |
| GTX 580 | -34 | -32 | -33 |
| GTX 570 | -42 | -40 | -41 |
| HD 7970 | -15 | -10 | -12 |
| HD 7950 | -30 | -25 | -27 |
| HD 7770 | -66 | -65 | -66 |
| HD 7750 | -75 | -74 | -75 |
| HD 6990* | 13 | 21 | 17 |
| HD 6990 | 9 | 16 | 13 |
| HD 6970 | -41 | -37 | -39 |
| HD 6950 | -49 | -45 | -47 |
| HD 6870 | -56 | -52 | -54 |
| HD 6850 | -66 | -63 | -64 |
| GTX 680OC | 21 | 15 | 18 |
| GTX 590OC | 8 | 10 | 9 |
| GTX 580OC | -20 | -17 | -18 |
| GTX 570OC | -32 | -28 | -30 |
| HD 7970OC | 13 | 20 | 16 |
| HD 7950OC | 0 | 5 | 3 |
| HD 7770OC | -60 | -58 | -59 |
| HD 6990OC | 23 | 30 | 26 |
| HD6970OC | -35 | -31 | -33 |
| HD 6950OC | -38 | -33 | -35 |
| HD 6870OC | -51 | -47 | -49 |
| HD6850OC | -57 | -53 | -55 |
| HD 7970x2 | 71 | 82 | 76 |
| HD 7950x2 | 41 | 50 | 46 |
| HD 7770x2 | -33 | -30 | -32 |
| HD 6990x2 | 126 | 94 | 110 |
Unigine Heaven
Benchmark version 2.5.
Settings:
- DirectX11;
- Stereo 3D - off;
- Shaders - high (high);
- Tessellation - normal (normal);
- Anisotropic filtering (Anisotropy) - 16x;
- Anti-aliasing - 8x.
advertising
Frames/sec
Min|Avg.1920x1080
Please enable JavaScript to see charts
2560x1440
Please enable JavaScript to see charts
Frames/sec
OverclockingMin|Avg.
1920x1080
Please enable JavaScript to see charts
2560x1440
Please enable JavaScript to see charts
| Model | Medium 1920% | Medium 2560% | The average |
| GTX 680 | 0 | 0 | 0 |
| GTX 590 | 14,4 | 23,9 | 19,1 |
| GTX 580 | -27,5 | -22,6 | -25,1 |
| GTX 570 | -39,7 | -41,3 | -40,5 |
| HD 7970 | -6,9 | 2,3 | -2,3 |
| HD 7950 | -19,3 | -11,3 | -15,3 |
| HD 7770 | -59,7 | -67,7 | -63,7 |
| HD 7750 | -67,6 | -74,2 | -70,9 |
| HD 6990* | 24,9 | 36,1 | 30,5 |
| HD 6990 | 21,2 | 32,6 | 26,9 |
| HD 6970 | -33,3 | -28,1 | -30,7 |
| HD 6950 | -40,1 | -34,5 | -37,3 |
| HD 6870 | -44,9 | -61,0 | -53,0 |
| HD 6850 | -54,3 | -65,8 | -60,1 |
| GTX 680OC | 17,8 | 21,6 | 19,7 |
| GTX 590OC | 21,7 | 31,6 | 26,7 |
| GTX 580OC | -15,2 | -10,0 | -12,6 |
| GTX 570OC | -31,3 | -30,6 | -31,0 |
| HD 7970OC | 21,2 | 32,3 | 26,7 |
| HD 7950OC | 16,7 | 27,4 | 22,0 |
| HD 7770OC | -54,3 | -66,1 | -60,2 |
| HD 6990OC | 33,9 | 45,2 | 39,5 |
| HD6970OC | -28,8 | -24,2 | -26,5 |
| HD 6950OC | -29,6 | -24,5 | -27,1 |
| HD 6870OC | -38,8 | -57,4 | -48,1 |
| HD6850OC | -45,9 | -62,3 | -54,1 |
| HD 7970x2 | 90,3 | 111,0 | 100,6 |
| HD 7950x2 | 64,0 | 81,9 | 73,0 |
| HD 7770x2 | -20,0 | -66,8 | -43,4 |
| HD 6990x2 | 139,5 | 164,8 | 152,2 |
Metro 2033
advertising
Settings:- DirectX11;
- Resolution - 1920x1080, 2560x1440;
- Full screen anti-aliasing - AAA;
- Anisotropic Filtering (AF) - 16x;
- The quality is very high;
- Enhanced depth of field - on;
- Tessellation - on.
Frames/sec
Min|Avg.1920x1080
Please enable JavaScript to see charts
2560x1440
Please enable JavaScript to see charts
Frames/sec
OverclockingMin|Avg.
1920x1080
Please enable JavaScript to see charts
2560x1440
Please enable JavaScript to see charts
| Model | Medium 1920% | Medium 2560% | The average |
| GTX 680 | 0 | 0 | 0 |
| GTX 590 | 29,6 | 37,4 | 33,5 |
| GTX 580 | -12,4 | -8,7 | -10,5 |
| GTX 570 | -24,4 | -23,0 | -23,7 |
| HD 7970 | 14,0 | 6,4 | 10,2 |
| HD 7950 | -0,9 | -1,9 | -1,4 |
| HD 7770 | -57,1 | -59,6 | -58,4 |
| HD 7750 | -63,9 | -64,2 | -64,0 |
| HD 6990* | 46,7 | 40,0 | 43,4 |
| HD 6990 | 43,1 | 36,2 | 39,7 |
| HD 6970 | -22,1 | -26,4 | -24,3 |
| HD 6950 | -30,5 | -33,2 | -31,8 |
| HD 6870 | -40,6 | -49,8 | -45,2 |
| HD 6850 | -50,1 | -57,4 | -53,7 |
| GTX 680OC | 19,6 | 19,6 | 19,6 |
| GTX 590OC | 37,2 | 46,4 | 41,8 |
| GTX 580OC | 1,6 | 6,0 | 3,8 |
| GTX 570OC | -12,0 | -10,9 | -11,5 |
| HD 7970OC | 42,7 | 37,7 | 40,2 |
| HD 7950OC | 34,1 | 34,0 | 34,0 |
| HD 7770OC | -52,4 | -53,2 | -52,8 |
| HD 6990OC | 58,5 | 49,8 | 54,1 |
| HD6970OC | -17,8 | -23,0 | -20,4 |
| HD 6950OC | -19,9 | -23,0 | -21,4 |
| HD 6870OC | -33,9 | -48,3 | -41,1 |
| HD6850OC | -39,7 | -52,5 | -46,1 |
| HD 7970x2 | 130,7 | 121,9 | 126,3 |
| HD 7950x2 | 103,6 | 100,0 | 101,8 |
| HD 7770x2 | -16,0 | -17,0 | -16,5 |
| HD 6990x2 | 132,3 | 158,9 | 145,6 |
Aliens vs Predator 3
Version - 1.03.
Settings:
- DirectX11;
- Resolution - 1920x1080, 2560x1440;
- Full screen anti-aliasing (MSAA) - 4x;
- Anisotropic filtering (AF) - 16x;
- Texture Quality - very high;
- Shadow Quality - high;
- Screen Space Ambient Occlusion - on;
- Tessellation (DX11 Tessellation) - on;
- Improved shadows (DX11 Advanced shadow sampling) - incl.
Frames/sec
Min|Avg.1920x1080
Please enable JavaScript to see charts
2560x1440
Please enable JavaScript to see charts
Frames/sec
OverclockingMin|Avg.
1920x1080
Please enable JavaScript to see charts
2560x1440
Please enable JavaScript to see charts
| Model | Medium 1920% | Medium 2560% | The average |
| GTX 680 | 0 | 0 | 0 |
| GTX 590 | 30,8 | 36,3 | 33,5 |
| GTX 580 | -15,7 | -13,4 | -14,5 |
| GTX 570 | -28,8 | -27,7 | -28,3 |
| HD 7970 | 6,0 | 9,1 | 7,6 |
| HD 7950 | -10,1 | -7,4 | -8,7 |
| HD 7770 | -59,4 | -59,4 | -59,4 |
| HD 7750 | -70,0 | -70,3 | -70,2 |
| HD 6990* | 74,9 | 76,9 | 75,9 |
| HD 6990 | 68,5 | 71,1 | 69,8 |
| HD 6970 | -7,4 | -6,9 | -7,1 |
| HD 6950 | -19,3 | -19,7 | -19,5 |
| HD 6870 | -30,9 | -30,3 | -30,6 |
| HD 6850 | -44,7 | -44,3 | -44,5 |
| GTX 680OC | 19,7 | 20,0 | 19,9 |
| GTX 590OC | 38,7 | 45,1 | 41,9 |
| GTX 580OC | -1,2 | 1,7 | 0,3 |
| GTX 570OC | -17,0 | -15,7 | -16,4 |
| HD 7970OC | 36,4 | 39,4 | 37,9 |
| HD 7950OC | 9,5 | 9,7 | 9,6 |
| HD 7770OC | -54,4 | -54,9 | -54,6 |
| HD 6990OC | 90,7 | 93,7 | 92,2 |
| HD6970OC | -0,4 | -1,1 | -0,8 |
| HD 6950OC | -3,3 | -4,0 | -3,6 |
| HD 6870OC | -23,6 | -22,9 | -23,2 |
| HD6850OC | -31,9 | -34,0 | -33,0 |
| HD 7970x2 | 125,7 | 132,6 | 129,1 |
| HD 7950x2 | 89,7 | 95,1 | 92,4 |
| HD 7770x2 | -19,3 | -20,9 | -20,1 |
| HD 6990x2 | 231,1 | 234,0 | 232,6 |
On March 22, 2012, a super-new product from the company was presented NVIDIA– architecture graphics card Kepler – NVIDIA GeForce GTX 680. Another technological breakthrough from engineers has brought more energy efficiency, more performance. Of course, more money from the population. I can evaluate the capabilities of a new generation card using the example ASUS GeForce GTX 680.
The ASUS video card comes in a cardboard box, the background of which is designed as an aged metal, torn apart by the claws of an unknown predator. The icon reminds us that the proprietary GPU Tweak utility will come to the rescue if you want to overclock and fine-tune the video card right in real time. In addition, ASUS talks about 2 gigabytes of video memory, support for DirectX 11 and PCI-Express 3.0 bus.
The reverse side, as usual, is left to describe proprietary gadgets and technologies, including GPU Boost and adaptive vertical sync. GPU Boost is somewhat analogous to Intel technology, with the difference that processors increase the multiplier, and the video card adds frequency by just megahertz. With the help of programs, you can set a heat dissipation limit and try to overclock the video card higher than the Boost Clock frequency. Looking ahead a little, I’ll say that the GeForce GTX 680 is all right with overclocking.
The ASUS GeForce GTX 680 video card is a copy of the reference sample. The original solutions of engineers are used on other models, such as DirectCu II, which I hope will get to my tests.
The video core is powered by a four-phase power system. Video memory receives power in a two-phase scheme.
Additional power is supplied to the video card through two connectors PCI Express 6-pin. Their location is very unusual, one above the other, while the connector farthest from the board is recessed by a centimeter in height. Power consumption is 195 watts and the manufacturer recommends using at least a 550-watt power supply.
The video card supports the simultaneous connection of four monitors, for which it has four video outputs - two DVI, HDMI 1.4a and DisplayPort.
The core is cooled with a small heatsink. A complex design using three flat heat pipes copes with the temperature perfectly.
A separate metal plate covers the entire board and cools the video memory chips and power elements on the board. The entire system is blown with a turbine-type fan.
To test the performance of the system, the following system was assembled:
CPU |
|
Motherboard |
|
video card |
|
RAM |
|
Power Supply |
|
HDD |
|
Frame |
|
Monitor |
|
Keyboard |
|
Mouse |
|
A set of test applications with settings:
Synthetic tests:
- 3D Mark 11, Entry, Performance, Extreme presets
- 3D Mark Vantage, Performance preset
- Unigine Heaven HWBOT Edition, DX 11, Extreme preset
Game tests:
- Aliens vs Predator DX11 Benchmark, 1920*1080
- Dirt 3, 1920*1080, built-in benchmark
- Metro 2033, 1920*1080, built-in benchmark
- Lost Planet 2 DX 11, 1920*1080, benchmark A
- Street Fighter benchmark, 1920*1080
For testing, the ASUS GeForce GTX 680 video card was overclocked to 1220 MHz for the graphics core and 1600 MHz for the video memory. For comparison, we chose the flagship of the previous generation - Leadtek GeForce GTX 580, the current leader from the competing camp - AMD Radeon HD 7970 and the dual-chip card of the previous generation - AMD Radeon HD 6990.
Previously, 3D Mark Vantage was considered the patrimony of NVIDIA cards - those days are gone. The Reds are ahead by a serious margin.
3D Mark 11 was released quite recently and its results are more fair. In the nominal mode, the ASUS GeForce GTX 680 loses a little to its competitor in the face of the AMD Radeon HD 7970, but overclocking allows you to pull yourself up very close.
The Aliens vs Predator gaming test also favored red. The advantage of the GeForce GTX 680 over its predecessor is about 20%
Dirt3 unleashes its potential NVIDIA graphics cards- leadership is deserved and natural.
Lost Planet 2's love for NVIDIA graphics accelerators is known and confirmed.
Metro 2033 goes better on AMD Radeon, but the advantage is minimal, so we should rather talk about parity.
The effect of overclocking the ASUS GeForce GTX 680 is practically absent, which is somewhat incomprehensible. Perhaps some feature of the drivers or architecture.
The synthetic test favored Radeon, but keep in mind that AMD products disable tessellation, which is actively used in this test.
Final thoughts.
The ASUS GeForce GTX 680 video card demonstrates high performance and a significant advantage over the flagships of the previous generation based on the Fermi core. In the battle with red competitors, it is not always possible to be ahead, but this is compensated by lower power consumption and best technology smoothing. Expect even more from non-reference versions high frequencies and then the product will be even more successful. According to my personal assumptions, the potential of the GeForce GTX 680 is very high. ASUS will certainly be able to use it in the right direction. For now, for high quality and excellent performance ASUS GeForce GTX 680 wins Our Choice award from OCClub
The other day NVIDIA company introduced GeForce GTX 770, which is a slightly overclocked version of the former flagship of the six hundredth series. One could say that the days of the GeForce GTX 680 are already fully numbered and it's time for this fighter to retire. But in this review, we will make sure that it is too early to write off the old guard. Some GeForce GTX 680s may be more interesting to the buyer than the updated GeForce GTX 770.
Last year, we considered the top overclocker version of the GeForce GTX 680 from ASUS - GTX680-DC2T-2GD5. The current hero, ASUS GTX680-DC2O-2GD5, differs little from the familiar flagship. As we will see, these are practically twin brothers, with the exception of different factory frequencies and one letter in the name of the models.
Let's start with the traditional packaging inspection. The box is standard, large. On the front side there is an image of a cooler. A large inscription immediately informs us that the Boost Clock frequency has been increased to 1084 MHz.
The scope of delivery is as follows:
- power adapter from two PCI-E 6pin to PCI-E 8pin;
- SLI bridge;
- disk with software;
- instruction.
Outwardly, this is a complete copy of the already mentioned ASUS GTX680-DC2T-2GD5 video adapter. A large stiffening rib is screwed on the side.
Everything is designed in the corporate style of ROG - black color is combined with red accents.
The back side is covered with a metal plate. On the side of the textolite, several marked holes are visible. ASUS GTX680-DC2O-2GD5 supports VGA Hotwire technology, and when the card is connected through these pins to the corresponding lines of ASUS Republic of Gamers motherboards, it becomes possible to control and manage the voltage of the GPU and memory.
There are four connectors on the rear panel: DisplayPort, HDMI and two DVI.
For cooling, a DirectCU II cooler is used, which is already well known to our readers from other reviews of ASUS video cards. Five heatpipes with direct contact technology, a two-piece thin heatsink and very large fans with an actual impeller size of 92mm (which is actually 100mm). The uniqueness of this system is the use of 25 mm thick fans. Typically, other manufacturers install thin fans and use the remaining space to increase the heatsink area. ASUS's approach is the opposite, but such a system has long proven its effectiveness.
The radiator sat so tightly inside the metal casing that it completely refused to be removed after unscrewing all the fixing screws. We did not become particularly zealous because of the danger of damage to the radiator. If you are interested in more detailed photos, you can view them in the review. ASUS HD7970-DC2-3GD.
There is a mandatory VRM block heatsink.
The plate that covers the reverse side of the board does not remove heat from any components. It only serves as an amplifier of the overall design and at the same time performs protective functions.
No surprises present and inspection printed circuit board. The design completely copies ASUS GTX680-DC2T-2GD5. Eight-phase GPU voltage converter controlled by DIGI+ ASP1212 controller. The memory power subsystem is implemented using two phases.
The heart of the video card is the GK104-400-A2 graphics chip.
Soldered memory chips Samsung K4G20325FD FC03.
Factory overclocking seems to be modest. The base frequency of the chip has been increased from 1006 MHz to 1020 MHz. The memory does without overclocking and operates at an effective 6008 MHz.
According to GPU-Z, the ASIC Quality parameter is 80.8%.
The real core frequency is stable at around 1124 MHz. Benchmark Crysis: Warhead at maximum graphics quality only warmed up the GPU to 66 ° C. The fan speed barely exceeded 1500 rpm, the noise was low.
In the 12-minute Unigine Valley benchmark, the peak temperature did not exceed 68 °C. The fans were already spinning up to 1620 rpm.
Acceleration without exaggeration is magnificent. The card took 1200 MHz at the base frequency, which gave a real Boost frequency of 1293 MHz. Our copy did not reach the level of ASUS GTX680-DC2T-2GD5 by some 15 MHz. But the GTX680-DC2O-2GD5 turned out to be much better in memory overclocking, overcoming 7750 MHz. An incredible result for the GeForce GTX 680! We have not achieved this on any map of this series. It should be noted that this is the only GeForce GTX 680 based on Samsung K4G20325FD FC03 chips. But exactly the same chips were equipped with MSI N770 TF 2GD5/OC, which overcame even more impressive 8002 MHz.
Acceleration is excellent. And if you resort to flashing a modified BIOS with increased voltage, then the results of GPU overclocking will be even more impressive. In this case, we limited ourselves to simple overclocking at native voltage, and for the corresponding instructions on updating the BIOS, you can refer to our old article.
Turnover had to be raised. We fixed them at around 43%, which corresponded approximately to 1900 rpm. The noise was noticeable, but tolerable. The core temperature in Crysis: Warhead did not exceed 69 ° C.
The Unigine Valley benchmark warmed up the card to 72°C. AT this application frequency drops up to 1280 MHz have already been observed, although very rare.
The table below shows the official specifications of the test participants. Performance comparison charts show the full operating frequency range for the GeForce, including barely-reaching peaks.
| Video adapter | GeForce GTX 780 | GeForce GTX 770 | GeForce GTX 680 | Radeon HD 7970 GHz Edition | Radeon HD 7970 | |
| Core | GK110 | GK104 | GK104 | GK104 | Tahiti | Tahiti |
| Number of transistors, million pieces | 7100 | 3500 | 3500 | 3500 | 4312 | 4312 |
| Process technology, nm | 28 | 28 | 28 | 28 | 28 | 28 |
| Core area, sq. mm | 561 | 294 | 294 | 294 | 365 | 365 |
| Number of stream processors | 2304 | 1536 | 1536 | 1536 | 2048 | 2048 |
| Number of texture blocks | 192 | 128 | 128 | 128 | 128 | 128 |
| Number of render units | 48 | 32 | 32 | 32 | 32 | 32 |
| Core frequency, MHz | 863-900 | 1046-1085 | 1020-1084 | 1006-1059 | 1000-1050 | 925 |
| Memory bus, bit | 384 | 256 | 256 | 256 | 384 | 384 |
| Memory type | GDDR5 | GDDR5 | GDDR5 | GDDR5 | GDDR5 | GDDR5 |
| Memory frequency, MHz | 6008 | 7010 | 6008 | 6008 | 6000 | 5500 |
| Memory size, MB | 3072 | 2048 | 2048 | 2048 | 3072 | 3072 |
| Supported version of DirectX | 11.1 | 11.1 | 11.1 | 11.1 | 11.1 | 11.1 |
| Interface | PCI-E3.0 | PCI-E3.0 | PCI-E3.0 | PCI-E3.0 | PCI-E3.0 | PCI-E3.0 |
| Declared power level, W | 250 | 230 | n/a | 195 | 250 | 250 |
test bench
The test bench configuration is as follows:
- Processor: Intel Core i7-3930K (3, [email protected].4 GHz, 12 MB);
- cooler: Thermalright Venomous X;
- motherboard: ASUS Rampage IV Formula/Battlefield 3 (Intel X79 Express);
- memory: Kingston KHX2133C11D3K4/16GX (4x4 GB, [email protected] MHz, 10-11-10-28-1T);
- system disk: Intel SSD 520 Series 240GB (240 GB, SATA 6Gb/s);
- secondary drive: Hitachi HDS721010CLA332 (1 TB, SATA 3Gb/s, 7200 rpm);
- power supply: Seasonic SS-750KM (750 W);
- monitor: ASUS PB278Q (2560x1440, 27”);
- operating system: Windows 7 Ultimate SP1 x64;
- GeForce driver: NVIDIA GeForce 314.22;
- GeForce GTX 780/770 driver: NVIDIA GeForce 320.18;
- Radeon driver: ATI Catalyst 13.3 beta 3.
Test results
AT Assassin's Creed 3 our hero is almost as good as the GeForce GTX 770 in nominal terms and easily outperforms it when overclocked.
The advantage over the GeForce GTX 680 is minimal. When overclocked, ASUS leaves behind not only the reference, but also the boosted GeForce GTX 770.
ASUS shows a minimal difference with the reference in . When overclocked, it takes first place among all cards based on the GK104. The gap from the Radeon HD 7970 GHz Edition is at the level of 8-12%.
AT Warhead a meager advantage over a simple GeForce GTX 680. With an increase in frequencies, ASUS outperforms the accelerated GeForce GTX 770 by 3-4%. At higher frequencies, ASUS catches up with the nominal GeForce GTX 780.
In a simple GeForce GTX 680, it is slightly inferior to the Radeon HD 7970 GHz Edition, but the ASUS GTX680-DC2O-2GD5 is already slightly better. When overclocked, our hero is comparable to the nominal GeForce GTX 780, even a little faster.
ASUS's traditionally modest advantage over its simple brother in . When overclocked this time, they also differ slightly. Our hero is no longer able to compete with the older GeForce GTX 780.
the last part Hitman sympathizes with AMD solutions and no amount of overclocking helps the GeForce GTX 680 to catch up with the older Radeon. Among the trio, the GK104-based card is the best overclocked by ASUS.
Metro 2033
power usage
A system with ASUS consumes more watts than configurations with a regular GeForce GTX 680. The result is quite natural. Given the good overclocking potential, this indicates an initially increased voltage. But ASUS is still more economical than Radeon.
findings
The reviewed model GTX680-DC2O-2GD5 is a complete copy of the TOP version of the GeForce GTX 680 from ASUS. The initial frequencies are lower, but this is easily compensated by overclocking, and there is room for overclocking. The potential of the GTX680-DC2O-2GD5 is excellent. And this applies to both the capabilities of the GPU and memory. The considered video adapter became the record holder for overclocking video memory among all the representatives of the series that we had previously, only slightly losing in this regard to the GeForce GTX 770. As a result, you can easily get 25% higher performance entry level GeForce GTX 680. Cooling is also not up to par. The fact that this ASUS is quiet and cold at face value is immediately clear. For us, it is more important that the cooler easily copes with its task at an average speed even during overclocking. Frankly, we set the fan speed with a "margin" so that there were no problems with overheating. With more fine tuning it will not be difficult to achieve complete acoustic comfort at maximum frequencies. An argument in favor of the considered model is the fact that ASUS announced the GeForce GTX 770 so far only with two-slot DirectCU cooling. And it seems that there will be no version with a three-slot cooler at all. So the GTX680-DC2O-2GD5 does not lose its relevance at all. This video adapter not only deserves the title of one of the best in the GeForce GTX 680 pantheon, it also looks great against the background of the newly-minted GeForce GTX 770.
Testing equipment was provided by the following companies:
- ASUS- ASUS monitor PB278Q, ASUS Rampage IV Formula/Battlefield 3 board, GTX680-2GD5 and GTX680-DC2O-2GD5 graphics cards;
- MSI - MSI N770 TF 2GD5/OC and R7970 Lightning video cards;
- Intel- Intel processor Core i7-3930K and Intel SSD 520 Series 240GB;
- Kingston - Kingston KHX2133C11D3K4/16GX memory kit;
- Syntex - Seasonic SS-750KM power supply;
- Thermalright - Thermalright Venomous X.