Families of video cards AMD (ATI) Radeon Reference information. Testing AMD Radeon HD6800 series video cards Radeon hd 6800 series specifications

Specifications of reference cards based on R9XX family chips

Details: Cayman, Radeon HD 6900 series

• Chip code name "Cayman"
• 40 nm technology
• 2.64 billion transistors (nearly a quarter more than Cypress and 1.5 times more than Barts)
• Crystal area 389 mm2 (one and a half times larger than Barts)
• Core clock up to 880 MHz (for Radeon HD 6970)
• 24 SIMD cores including 384 stream processors and a total of 1536 scalar floating point ALUs (integer and float formats, support for IEEE 754 FP32 and FP64 precision)
• 24 large texture units, with support for FP16 and FP32 formats
• 96 texture address units and the same number of bilinear filtering units, with the ability to filter FP16 textures at full speed and support for trilinear and anisotropic filtering for all texture formats
• 32 ROPs with support for anti-aliasing modes with the possibility of programmable sampling of more than 16 samples per pixel, including with FP16 or FP32 framebuffer format. Peak performance up to 32 samples per clock (including for FP16 buffers), and in colorless mode (Z only) - 128 samples per clock

Radeon HD 6970 Graphics Specifications

• Core clock 880 MHz
• Number of universal processors 1536
• Number of texture blocks - 96, blending blocks - 32
• Effective memory frequency 5500 MHz (4×1375 MHz)
• Memory type GDDR5
• Memory capacity 2 gigabytes
• Memory bandwidth 176 gigabytes per second.
• Theoretical maximum fill rate is 28.2 gigapixels per second.
• Theoretical texture fetch rate is 84.5 gigatexels per second.
• Two CrossFireX connectors
• PCI Express 2.1 bus
• Power Consumption 20W to 250W (typical gaming power consumption up to 190W)
• One 8-pin and one 6-pin power connector
• Dual slot design
• MSRP for the US market $369

Radeon HD 6950 Graphics Specifications

• Core clock 800 MHz
• Number of universal processors 1408
• Number of texture units - 88, blending units - 32
• Memory type GDDR5
• Memory capacity 2 gigabytes
• Memory bandwidth 160 gigabytes per second.
• Theoretical maximum fill rate is 25.6 gigapixels per second.
• Theoretical texture fetch rate is 70.4 gigatexels per second.
• Two CrossFireX connectors
• PCI Express 2.1 bus
• Connectors: DVI Dual Link, DVI Single Link, HDMI 1.4a, two mini DisplayPort 1.2
• Power Consumption 20W to 200W (typical gaming power consumption up to 140W)
• Two 6-pin power connectors
• Dual slot design
• MSRP for the US market $299

The use of the proven 40-nanometer process technology nevertheless allowed AMD to release a new top-end GPU, albeit not in the same form as it could be at 32 nm. The complexity of Cayman has grown by less than a quarter compared to Cypress, as has the core area, but some characteristics that affect performance have remained almost the same. This is the number of ALUs, and the same number of ROPs, and the bandwidth of the video memory has not grown much. But still, thanks in large part to the increased clock speeds and increased efficiency of the new AMD chip, it should outperform Cypress on average.

The principle of naming models has been somewhat changed from the previous generation. Compared to the previous series, the top solutions have changed not only the first, but also the second digit of the index. The Radeon HD 6970 and HD 6950 are the most productive single-chip solutions and should replace the HD 5870 and HD 5850 video cards, becoming higher in the lineup than the recently released HD 6800 family solutions. As for the comparison with a competitor, at the above recommended prices, it is clear that in terms of HD performance The 6970 is at the same level or somewhat more productive than the GeForce GTX 570, but the HD 6950 has a competitor on a different chip - the GTX 560 Ti.

The two versions of the series, as is customary with AMD video cards, differ both in the clock frequencies of the video chip and memory, and in the disabled part of the execution units in the younger model. Both video cards of the new series are equipped with GDDR5 memory of the same size of 2 gigabytes. The optimal amount of memory for today is still 1 gigabyte, but it is quite possible that for top models this amount is justified, since in some cases a shortage of 1 GB of memory will still be observed, and even for games on three monitors (Eyefinity) a screen buffer of this size would be very useful. By the way, the company's partners have already released the Radeon HD 6950 model with 1 GB of video memory at a lower cost.

Both video cards have a two-slot cooling system, which is covered with a plastic shroud, familiar to all modern AMD boards, along the entire length of the card. The power consumption of the younger card is lower, which made it possible to get by with two 6-pin power connectors in its case. In addition to the maximum power consumption, AMD now also indicates typical gaming power - a consumption indicator measured during testing in a set of 25 popular games.

Cayman architecture

When designing Cayman (namely, this is the code name received by the company's new GPU), the main tasks of AMD engineers were to create an efficient graphics and computing architecture with new GPGPU capabilities, a significant increase in the performance of geometric blocks, improvements in algorithms that affect rendering quality (texture filtering and full-screen anti-aliasing ), as well as improved power management.

Apparently, the Cayman architecture can be called an intermediate solution between the Cypress architecture and the never-born 32-nanometer architecture, since only some of its features were included in the new GPU. Interestingly, the Cayman sizing engineers goal was +15% Cypress footprint, allowing those extra transistors to be spent on some of the new compute and graphics capabilities we'll cover below. So, let's see what happened with AMD.

When looking at the chip's scheme, two blocks for processing geometry and tessellation immediately attract attention (graphics engine, including a rasterizer, tesselator and some other blocks), as well as a dual dispatcher. This is one of the most important innovations in Cayman, which was clearly prompted by the lag in geometry processing speed from a competitor that has had a parallelized graphics pipeline for almost a year.

The most important architectural change was the superscalar VLIW4 architecture of computing processors, in contrast to the previous VLIW5. On the one hand, this may seem like a deterioration, because each of the available processors can now perform fewer operations in parallel. But on the other hand, this can increase the efficiency of use (efficiency) of stream processors, since picking up four independent commands is clearly easier than five.

In total, the new graphics processor includes 24 SIMD cores, each of which consists of 16 processors that can calculate up to four instructions simultaneously. In other words, the total number of computing units in Cayman is 24×16×4=1536 pieces, which is even somewhat less than in Cypress. But since the efficiency of using these blocks should obviously increase, then the performance will also increase, most likely.

Each SIMD core of the new GPU has four texturing units, as in previous GPUs, that is, the total number of texture processors is 96 TMUs. This is somewhat more than Cypress, and noticeably more than the competitor's top-end chip. Thus, the advantage in texturing should remain with AMD. Other numerical characteristics differ little from the same HD 5800 and HD 6800, the chip has four 64-bit memory controllers and a 256-bit bus as a whole, as well as 32 ROPs. Although they are still different from those used in previous GPUs, and this will be discussed later.

Stream processor architecture

New stream processors differ from previous ones in that they can execute up to four independent instructions simultaneously (4-way co-issue), and all four ALUs in the processor have the same capabilities, unlike the previous architecture. Recall that each Cypress stream processor has four ALUs + a special-purpose SFU (also called a “T-unit”) to perform transcendental functions (sine, cosine, logarithm, etc.), and Cayman executes such commands when three of the four "regular" ALUs.

All together, this theoretically gives a better indicator of the efficiency of the use of stream processors, compared to VLIW5. Although the VLIW5 provides reasonably high efficiency in many cases, the average ALU utilization is well below 100%, and often only three or four out of five are occupied. Reducing the number of ALUs in each processor increases their efficiency, and according to AMD, the improvement in the ratio of computing speed and chip area is about 10%. Plus, an additional bonus is the simplification of control blocks: scheduler and register management.

Another important detail of the transition from VLIW5 to VLIW4 is that it is more difficult for an asymmetric architecture to optimize and compile efficient code. And for a symmetrical VLIW4 block, the compiler's work is simplified. And in this we see the still undiscovered potential of Cayman - most likely, the compiler is not optimized enough for the new GPU yet, and in the future gains are very likely as the compiler is optimized for the new architecture.

The new VLIW4 architecture has resulted in an increase in double precision performance. 64-bit calculations are now only four times slower than 32-bit ones. And for solutions of the previous architecture, this ratio was lower - 1/5. This change made it possible to increase the peak performance of 64-bit computing of the new Radeon HD 6970 to 675 GFLOPS (for comparison, this figure is 544 GFLOPS for the HD 5870).

ROP block changes

The ROPs in AMD's new chip also received some improvements. Cayman is now able to process data significantly faster in some formats, including 16-bit integer (twice as fast) and one- or two-component 32-bit (two to four times faster, depending on the number of components). This improvement is most important for the now widespread cases of deferred rendering, although the use of 32-bit buffers in games is still clearly limited.

Non-graphical computing on the GPU

Perhaps the biggest change to the Cayman has been in terms of computing power. First of all, it is necessary to note the asynchronous sending of commands for execution and the simultaneous execution of several computing processes (kernel), each of which has its own command queue and its own protected virtual memory area. In fact, Cayman introduced the possibility of computing on the principle of MPMD (Multiple Processor/Multiple Data) - when several processors execute many data streams.

Previous AMD architectures had the ability to run and distribute multiple processes (kernel) at the same time, but they had only one instruction pipeline, which made it difficult for compute and graphics applications to run simultaneously. The new GPU architecture is capable of efficiently executing multiple instruction streams simultaneously. Threads have their own separate ring buffers and queues, and the execution order of commands is independent and asynchronous, and they are executed depending on priority. This allows you to run calculations and get the final result out of turn.

Also, for each kernel, the new chip provides independent virtual memory, and all command streams are now protected from each other. And in addition to asynchronous command delivery, the chip has two bidirectional direct memory access (DMA) controllers to help increase throughput in both directions.

But that's not all "computational" changes in the Cayman. It became possible to fetch data from memory bypassing the ALU directly to local memory, and optimized reading and combined data writing increased the performance of the I / O subsystem. Also in the new GPU, flow control has been improved and much more.

Parallel geometry processing

In our materials, we have repeatedly mentioned that one of the main architectural advantages of competing solutions from NVIDIA is parallelized geometry processing, which is used in all their modern solutions, which are very effective when using tessellation. Geometric primitives in AMD's top-end chips are processed in 16 blocks simultaneously, as opposed to one block in Cypress and Barts, as well as in other previous chips.

Accordingly, AMD urgently needed to improve the performance of geometric blocks. A partial step was taken back in Barts, the optimizations of which led to an increase in the speed of geometry processing and tessellation by one and a half times at best. But even the seventh generation tessellator was still seriously inferior to the first generation Fermi tesselators.

The geometry and tessellation blocks in the Cayman are now called the eighth generation, and they received a double-speed geometry setup, improved geometry data buffering, and a double geometry processing block. That's right, AMD also had to parallelize the work on geometric data, although not as radically as it is done in the competitor's GPU.

The double geometry block in Cayman processes two primitives per cycle, that is, the speed of transformation and discarding of the back faces (backface culling) has doubled, and the load between the blocks is distributed using tiling. Together with improved buffering, according to AMD, this leads to an increase in tessellation performance in the top-end Radeon HD 6970 solution by up to three times compared to the HD 5870.

But still, as you can see, most often the speed of processing geometry and tessellation has doubled, and not tripled. Even according to AMD itself. By the way, they also give figures from games and benchmarks using tessellation, and the gains there reach impressive figures of the order of 30-70%, depending on the number of tessellated surfaces and the degree of fragmentation of primitives. We will check these figures in the next part of the material, dedicated to the performance studies of new solutions in synthetic tests and some of the gaming ones that also use tessellation.

One of the goals of the new architecture was to improve the quality of rendering. This concerns both the improvement of existing algorithms for texture filtering and anti-aliasing, and the emergence of new features, such as a new type of full-screen anti-aliasing - morphological (MLAA - MorphoLogical Anti-Aliasing).

Some of the new features are also available on the younger representatives of the series - Radeon HD 6800 video cards, but there is one hardware innovation that appeared in the HD 6900 series, in the Cayman chip. This is an improved full screen anti-aliasing method called Enhanced Quality Anti-Aliasing (EQAA). In short, this is an analogue of Coverage Sampling Anti-Aliasing (CSAA), which NVIDIA has had since the time of the G80 chip (GeForce 8800 series), which we talked about a few years ago.

The essence of the method is that the colors of the samples and the depth are stored separately from information about their location, and there can be 16 samples per pixel with 8 calculated depth values, which saves bandwidth. The method avoids passing and storing one color or Z value for each subpixel, refining the average value of the screen pixel due to more detailed information about how this pixel overlaps the edges of the triangles. The following picture will make it easier for you to understand this confusing explanation:

In previous AMD chips (including the HD 6800 series), the number of calculated and stored samples was the same. In HD 6900 series solutions, these two values ​​can be changed independently of each other, and the number of samples per pixel and the number stored in the buffer can be different. This allows you to get a quality higher than conventional multisampling (MSAA) while maintaining a relatively high performance.

EQAA makes it possible to provide anti-aliasing quality much higher than that of MSAA 4x, with only a slight loss in performance. According to AMD, the difference in performance between EQAA enabled and disabled modes in games is a few percent, which correlates well with the results of NVIDIA video cards.

An additional positive factor is that the method is compatible with adaptive anti-aliasing (Adaptive AA), super-sampling (Super-Sample AA) and morphological anti-aliasing, which we talked about in the article about the Radeon HD 6800. But how is this EQAA turned on? AMD adopted the competitor's experience here too, introducing similar options for changing the anti-aliasing method into the driver settings (for example, from regular MSAA to EQAA, but not necessarily that way).

We discussed other rendering quality improvements in the new AMD solutions in the article about the Radeon HD 6800 family, as well as about "morphological" anti-aliasing and texture filtering improvements. Morphological Anti-Aliasing is a new anti-aliasing method known to us from some multi-platform games. This is a post-processing filter applied to the final image using a compute or pixel shader.

This method smoothes all the pixels in the scene, not just the edges of polygons and translucent textures like MSAA, and therefore, after it, excessive blurring of the picture can be noted. But this method is theoretically faster than supersampling, since it processes only the necessary areas where the filter has found sharp color transitions. The difference from another method known as edge-detect CFAA is that the filter is applied to all faces, not just the edges of triangles.

All these methods can be mixed with each other. In other words, EQAA is fully compatible with both so-called "custom resolve" filters and "morphological" anti-aliasing, and all of them can be applied simultaneously. This will improve the quality of rendering in case of excess performance, often found in top-end video cards.

AMD PowerTune technology

One of the most interesting changes in the Cayman, not directly related to 3D graphics, is a technology called PowerTune. Actually, things have been going on for a long time towards flexible control of the clock frequency, voltage and power supply of the GPU. The same central processors have long been able to smoothly or stepwise change performance and “gluttony”, reducing some parameters in idle time and increasing them under load. Yes, and video chips are also able to change the specified parameters, but until now they did it in steps and had no limits beyond which it would be impossible to go.

Regular games and other applications that use GPU computing rarely have high power requirements and do not approach dangerous power limits beyond the system's capacity. Unlike stability tests like Furmark and OCCT, which squeeze everything out of the system. Even in the Evergreen family (Radeon HD 5000 series) there was a certain rudimentary performance limiter when a certain consumption level was exceeded, and in the HD 6900 this system moved to a qualitatively different level.

The new GPU has special sensors in all blocks of the chip that monitor the load parameters, so the GPU constantly measures the load and power consumption and does not allow the latter to go beyond a certain threshold, automatically adjusting the frequency and voltage so that the parameters remain within the specified heat pack. This technology helps to set high GPU frequencies and at the same time not be afraid that the video card will go beyond the safe limits for power consumption. AMD provides the following applications as an example:

As you can see, the most demanding 3D applications are stability testing tools and some of the synthetic tests. But games, even the heaviest ones, do not require maximum energy from the GPU at all and do not go beyond the established limits.

Unlike earlier power management technologies, PowerTune provides direct control over the power consumption of the GPU, as opposed to indirect control by changing frequencies and voltages. And you no longer need to set a limiter for selected applications, the technology will work with the same success for all programs, including future ones.

For AMD, the technology is useful for several reasons at once: it will protect video cards from failure in some cases (for example, negligent and inattentive overclockers) and will allow you to squeeze the maximum performance out of the GPU without problems with power and cooling. It is also important that this technology allows the user to limit consumption using AMD OverDrive tools, as shown in the screenshot:

Naturally, the maximum consumption parameter can be adjusted only within certain limits and with the shifting of responsibility onto the shoulders of the user and depriving the latter of any guarantees. In some cases, it will be useful not only to increase this limit, but also to lower it, achieving a reduction in consumption in the absence of a need for high performance.

The change in GPU clock frequency and the resulting performance at different levels of maximum consumption are clearly displayed in the following graph. It shows the change in the GPU frequency of the Radeon HD 6950 video card in the Perlin Noise test from the 3DMark Vantage set in three modes: by default and with an increased power limit by 5% and 10%. This graph corresponds to what would happen when running the most power-hungry applications:

In the default mode, the GPU cannot run at 800 MHz all the time without exceeding the consumption limit set by AMD, and shows a result of 140 FPS. By adding 5% to the maximum power, the GPU frequency gets higher, but still often falls short of the maximum 800 MHz, resulting in 155 FPS. In the case of 10% added to the consumption limit, the chip always operates at a frequency of about 800 MHz and does not reach the changed consumption limit, while showing 162 average frames per second.

If we consider the reverse situation, when it is necessary to reduce consumption, then the technology will be useful in this case. AMD gives an example of Aliens vs Predator and three modes: default, -10% of max consumption, and -20%. If in the default and -10% modes the difference turned out to be small, then in the latter case, with a decrease in consumption by 30 W, you can get quite comfortable 40 FPS instead of 50 FPS at maximum consumption:

Thus, each user can customize PowerTune for themselves (subject to the waiver of guarantees, of course) and choose either lower system power consumption or higher performance in those applications in which the GPU becomes very demanding on power. You can even manually adjust lower consumption for continuous operation and maximum consumption for demanding applications.

Other changes

Among other interesting differences between video cards of the top Radeon HD 6900 family, I would like to note the following useful feature - the presence of two BIOS chips on the card and overwrite protection for one of them, which has factory settings. To do this, a microswitch is located on the board next to the CrossFire connectors.

The BIOS switch is used to ensure the operability of the video card in case of any problems the user encountered during the flashing process. This switch determines which image the video card will boot from: 1 - non-write-protected BIOS chip with the possibility of custom flashing, 2 - non-rewritable BIOS copy with factory settings.

This functionality is also designed to help in solving the problems of failed video cards. After all, now even in the event of an unsuccessful attempt to flash the BIOS, the user can always use the second way. One can only praise AMD for such a solution to user problems. Finally, it will be possible to throw away a spare PCI video card, carefully stored by many enthusiasts for such cases.

All of AMD's new family of graphics cards, both the HD 6800 and HD 6900, support DisplayPort 1.2 as part of AMD's enhanced Eyefinity Multi-Display Technology. Its difference from the previous ones is the ability to output several channels at once through one DisplayPort connector, which allows (more precisely, will allow in the future) to connect more monitors to one video card. To connect multiple monitors using one connector, you will need a special hub, purchased separately.

Cayman also contains a new video processing unit Unified Video Decoder 3, the most interesting new feature of which we see is the appearance of support for hardware decoding of the DivX / XviD format, which has not previously been accelerated on the GPU. But it's not just the decoding of this format that's improved in UVD3, it also now decodes MPEG-2 entirely on the GPU and supports dual-stream codecs to play 3D Blu-ray discs.

You can read more about the changes in display technologies, including Eyefinity capabilities, AMD HD3D technologies and the new generation of the Unified Video Decoder 3 video processing unit, in the theoretical review of the Radeon HD 6800 family solutions.

Details: Barts, Radeon HD 6800 series

• Chip codename "Barts"
• 40 nm technology
• 1.7 billion transistors (more than a quarter less than Cypress)
• Unified architecture with an array of common processors for streaming processing of multiple types of data: vertices, pixels, and more.
• Hardware support for DirectX 11, including a new shader model - Shader Model 5.0
• 256-bit memory bus: four 64-bit wide controllers with GDDR5 memory support
• Core frequency up to 900 MHz
• 14 SIMD cores including 1120 floating point scalar ALUs (integer and float formats, supports IEEE 754 FP32 precision)
• 14 large texture units, with support for FP16 and FP32 formats
• 56 texture address units and the same number of bilinear filtering units, with the ability to filter FP16 textures at full speed and support for trilinear and anisotropic filtering for all texture formats
• 32 ROPs with support for anti-aliasing modes with the possibility of programmable sampling of more than 16 samples per pixel, including with FP16 or FP32 framebuffer format. Peak performance up to 32 samples per clock (including for FP16 buffers), and in colorless mode (Z only) - 128 samples per clock
• Integrated support for RAMDAC, six Single Link or three Dual Link DVI ports, plus HDMI 1.4a and DisplayPort 1.2

Radeon HD 6870 Graphics Specifications

• Core clock 900 MHz
• Number of universal processors 1120
• Number of texture units - 56, blending units - 32
• Memory type GDDR5
• Memory size 1024 megabytes
• Theoretical maximum fill rate is 28.8 gigapixels per second.
• Theoretical texture fetch rate is 50.4 gigatexels per second.
• CrossFireX support
• PCI Express 2.1 bus
• Connectors: DVI Dual Link, DVI Single Link, HDMI 1.4a, two mini DisplayPort 1.2
• Power consumption 19 to 151 W (two 6-pin power connectors)
• Dual slot design
• US MSRP $239

Radeon HD 6850 Graphics Specifications

• Core clock 775 MHz
• Number of universal processors 960
• Number of texture units - 48, blending units - 32
• Effective memory frequency 4000 MHz (4×1000 MHz)
• Memory type GDDR5
• Memory size 1024 megabytes
• Memory bandwidth 128.0 gigabytes per second.
• Theoretical maximum fill rate is 24.8 gigapixels per second.
• Theoretical texture fetch rate is 37.2 gigatexels per second.
• CrossFireX support
• PCI Express 2.1 bus
• Connectors: DVI Dual Link, DVI Single Link, HDMI 1.4a, two mini DisplayPort 1.2
• Power consumption 19 to 127 W (one 6-pin power connector)
• Dual slot design
• US MSRP $179

The use of the same 40-nanometer process technology, but in a mature form, allowed AMD to release mid-range solutions that roughly correspond in performance to the previous top ones. The complexity of the chips has dropped by a quarter, as has the area of ​​the core, but many of the characteristics that affect performance have remained almost at the same level, largely due to increased clock frequencies. Naturally, the new chip has become even more energy efficient.

The principle of naming models has changed, we wrote about the reasons for this decision above. Compared to the previous series, both the first and second digits have changed. The Radeon HD 6870 and HD 6850 are designed to replace the HD 5870 and HD 5850, although they should be slightly slower in pairs. And the HD 6900 series cards became the new top models.

The two versions of the series, as usual for AMD video cards, differ in the clock frequencies of the video chip and memory, and the younger model also has some of the execution units disabled. Both video cards of the series are equipped with GDDR5 memory of the same size - 1 gigabyte. This is the optimal amount of memory for today, there will simply be no benefit from a larger amount on mid-range solutions.

And even the younger solution differs in the design of the board, and their reference coolers are different. Both video cards have a two-slot cooling system, covered by the usual plastic casing along the entire length of the card. But the power consumption of the younger card is lower, which made it possible to manage in its case with only one 6-pin power connector.

Architecture "Barts"

We reviewed the updated Cypress architecture in the corresponding background article. As you remember, there were no special changes in it, this is basically the development of the ideas of previous generations, although small modifications affected almost all blocks of the chip. And the differences between the Barts chip and Cypress are generally mostly quantitative, although not only.

So, what changes did the redesigned architecture bring to Barts? Basically, increased performance per watt and millimeter of area, that is, improved efficiency. Although AMD calls Barts "the second generation of DirectX 11", there are practically no changes in the architecture, they are almost exclusively quantitative - just a different number of execution units and a different balance between performance and consumption with cost.

Yes, some optimizations have resulted in faster geometry processing and tessellation, a sore point of AMD solutions, compared to competing ones. But these improvements did not change the speed of tessellation at times, but only one and a half to two times at best.

We find it more interesting to improve the quality of full-screen anti-aliasing and texture filtering, although they are more software rather than hardware. Support for decoding DivX and Blu-ray 3D video is also curious, and the improvements in AMD Eyefinity and support for the new HDMI 1.4a and DisplayPort 1.2 standards are very logical and timely.

Although these are mostly changes not related to the GPU core, but to other blocks that do not relate to the 3D part of the chip, which is the most interesting for us now. So, let's look at the block diagram of the new chip.

Let's see what has changed. In fact, these are only the blocks in the Graphics Engine and the total number of SIMD blocks. The tessellation block is now improved (this is the seventh generation, see below), there are two rasterizers (or the rate of processing primitives has been doubled, which is also quite likely), and the number of SIMD blocks has decreased from 18-20 (for Cypress) to 12-14 pieces ( at Barts), depending on the model.

The total number of stream processing processors has also decreased by the same amount, now there are a maximum of 1120 of them, in contrast to 1600 for Cypress. Everything else remains the same, and a 256-bit memory bus with support for GDDR5 video memory, and ROPs, and the rest.

Thanks to the higher clock speeds, the performance of the Radeon HD 6870 is higher than that of the HD 5850 (attention - lower than the HD 5870 even theoretically!), with a smaller GPU area. But this is a price comparison, and if we compare Barts and Cypress chips at the same frequency, then the solution announced today will be generally slower.

Tessellation and Geometry Processing

It is known that a relatively weak point of early AMD solutions was tessellation, which appears in DX11 applications. And it is quite logical that Barts partially corrected exactly this. The tessellation block in this GPU is already the seventh generation of the ATI/AMD tesselator (see slide below). The first appeared in the ancient ATI Radeon 8500, the second in the Xbox 360 console from Microsoft, and then came the series of AMD video cards. We will probably see the 8th generation already in the HD 6900 series ...

To be honest, we do not quite understand such a large number of generations of tessellators, especially if most of their changes were limited to the introduction of compatibility with DirectX versions, and even more so, exceptionally small performance gains. And you can also remember the solutions of a competitor, the very first generation of tessellators of which outperforms all the existing seven (or even eight) generations of AMD tesselators. So does it make sense to be proud of this figure?

However, more importantly, according to AMD's synthetic tests, the tessellation speed in the HD 6870 increased by 1.5-2 times compared to the HD 5870 (of course, we will check this in a practical study). Moreover, the new chip copes most effectively with medium levels of tessellation, and at high levels, the speed almost did not increase. But that won't be a problem, as games don't use these levels and won't need them anytime soon. Here is an example of increasing the complexity of the geometry at different degrees of partitioning:

This is already a pebble in the competitor's garden. Indeed, it is unlikely that anyone needs one-pixel triangles, and with too much detail, the efficiency of loading other blocks (rasterizers, for example) is significantly reduced, and in general, such work is not performed efficiently enough on current GPUs. The disadvantages of a high degree of tessellation are: extra work on shading (overshading), a large number of polygon edges that need to be processed during multisampling, etc. In general, this approach only causes a waste of resources, in the opinion of AMD representatives.

Ideally, you want to achieve the most efficient tessellated models, so that the size of each triangle is about 16 pixels per polygon. This is very beneficial for pixel-by-pixel processing, which is carried out by such blocks. This achieves the perfect balance between rendering quality and performance.

It is precisely to achieve this goal that methods such as adaptive tessellation serve, when high levels of splitting are used for objects in the foreground and individual surfaces that require high detail, and for distant objects, lower levels of tessellation are used, which improves performance and almost does not affect the quality of the final result. Pictures.

Improvements in rendering quality

As you know, the previous AMD chips made the right step towards achieving the highest quality picture - they now support a new anisotropic filtering algorithm, where the texture mip levels are arranged in perfect circles. You can also note the possibility of enabling anti-aliasing by supersampling, which significantly improves the overall quality of rendering.

What is pleasing, in the HD 6800 series, they continued to make changes aimed at improving the quality of the picture. On the one hand, almost everyone has already forgotten about it, since the quality of solutions from both AMD and NVIDIA is similar and generally quite good, but on the other hand, there is always room for improvement. In this case, AMD decided to introduce a new anti-aliasing mode, improve the quality of texture filtering, and (finally!) provide the ability to disable Catalyst AI optimizations.

The new anti-aliasing method is known from some multi-platform games Morphological Anti-Aliasing (MAA). This is not the anti-aliasing method we are used to, but rather a post-processing filter applied to the final image using a compute shader. This method smoothes all the pixels in the scene, not just the edges of polygons and translucent textures like MSAA, although it has the disadvantages of being too blurry, as you can see from the picture.

At the same time, MAA is faster than supersampling, since it processes only the necessary areas where sharp color transitions are found by the shader. The performance and essence of the algorithm is similar to the edge-detect CFAA method in AMD drivers, but anti-aliasing is applied to all sharp edges. More importantly, AMD Catalyst Control Center's MAA forcing method is promised to be compatible with all DirectX 9/10/11 applications.

But this new anti-aliasing method is a completely software innovation. What have AMD engineers changed in texture filtering algorithms? According to them, the anisotropic filtering algorithm has been redesigned to improve the processing of "noisy" textures, in particular, to obtain smoother transitions between texture mip levels with anisotropic filtering. At the same time, it is promised that there will be no loss in performance and no dependence of the quality of filtration on the angle of inclination of the surface, as it was before. In the screenshot, the HD 5800 is on the left and the HD 6800 is on the right.

Last but not least, there is a new user interface in the AMD Catalyst Control Center that allows you to change the quality of texture filtering and even completely disable all optimizations. To do this, a new Catalyst AI slider has been introduced into the driver settings:

As you can see, Texture Filtering Quality can have three values, and optimizations of texture formats are separately disabled (when one texture format is replaced in the driver by another, slightly lower quality, but faster), which AMD competitors had some complaints about.

Improvements in display technologies

It's useful to note that AMD's new DisplayPort 1.2 support is included in AMD's enhanced Eyefinity Multi-Display Technology. Its difference is the ability to output several channels at once through one DisplayPort connector, which will allow you to connect more monitors to one video card.

To connect multiple monitors using a single connector, you will need a special hub or daisy chain connection of monitors. DisplayPort 1.2 provides support for more monitors, higher resolutions and refresh rates, including the next generation of stereo monitors. By the way, all monitors can display images of different resolutions and refresh rates.

The new AMD graphics cards have an HDMI 1.4a port that is suitable for stereo output. This uses a special stereo frame transmission standard supported by new 3D TVs, so there will be no problems with stereo output on them (read a separate section on AMD's support for stereo rendering below).

An important factor in the quality of picture output is high-quality color correction when displaying images on monitors with an extended color gamut. And the AMD Radeon HD 6800 series has the right hardware engine for the task.

But multi-monitor technologies and image output technologies in general do not make much sense without proper support. And here everything is in order, there are already more than three dozen monitors with DisplayPort connectors on the market, and about fifty games specially optimized and prepared for multi-monitor output (and hundreds of other games are simply compatible with Eyefinity technology). Also, inexpensive DP to Single-Link DVI adapters have recently appeared, allowing you to connect several inexpensive monitors to one video card.

There are no less improvements in the drivers, in addition to everything that is already in the settings (dividing devices into groups, an advanced configurator, color correction for each device separately, display frame compensation, CrossFireX support, etc.), new modes will soon be added, such as a group of monitors 5 ×1 in portrait mode, automatic HydraGrid output, etc.

AMD HD3D technology

Seeing the successful promotion of stereo vision in the market, AMD could not stand aside without coming out with another open initiative. Now it belongs to stereo rendering. Announced at GDC 2010, the initiative is about collaborating between software and hardware vendors to provide a wide range of solutions, lower costs, and increase flexibility.

The initiative was supported by a large number of companies. For example, software for converting to Stereo 3D is produced by DDD and iZ3D, 3D video playback is handled by Cyberlink, Arcsoft, Roxio and Corel. Display manufacturers LG, Samsung, CMI and Viewsonic are responsible for the hardware, while the production of glasses and transmitters remains with Bit Cauldron, XpanD and RealD.

Actually, the Stereo 3D initiative does not offer anything new, it's all the same stereo monitors and stereo glasses, stereo games and support for Blu-ray 3D, software for converting content to stereo format, etc. AMD sees its task in providing the capabilities of AMD HD3D technology for games in stereo. To do this, video drivers provide support for 4-buffer rendering in DirectX 9, DirectX 10 and DirectX 11 applications, and with the help of partners from DDD and iZ3D, more than 400 games in stereo are already supported.

So, TriDef 3D Experience from DDD allows you to view photos and videos in stereo format, TriDef Ignition automatically "converts" about four hundred DirectX 9, 10 and 11 games to stereo format, and TriDef Media Player does the same with video data from DVD and high-definition video. permissions. Moreover, it is stated that the first stereo solutions based on AMD Radeon HD were shown (where and to whom is a separate question) a year ago, in October 2009. This solution is compatible with all stereo image output standards, all types of stereo glasses and "glassless" technologies.

By the way, about glasses. Colin Baden, CEO of Oakley, world-famous for sports optics and sunglasses, spoke at the AMD media event. He spoke about the Oakley HDO-3D stereo glasses model. Naturally, not without boasting, these glasses were called “the first optically correct stereo glasses on Earth”, allegedly reducing the effects of flare and ghosting of the picture, noticeable in many cases, including when using glasses from the 3D Vision kit. It would be interesting to compare these options live, but for now it remains to believe (or not believe) on the word.

By the way, AMD plans to launch a portal dedicated to HD3D stereo output technology on the website soon, helping users to get information about software and hardware solutions for games, viewing photos and videos in stereo format. With due diligence and funds, it can turn out well.

Unified Video Decoder 3 video processing unit

Radeon solutions have long been renowned for their video decoding and processing capabilities. Since the days of ATI, it was they who had some of the best solutions in this area. Subsequently, AMD continued these traditions. UVD3 has not only support for decoding new formats, but also better post-processing of video data.

New post-processing capabilities have led to a further strengthening of positions in the well-known test HQV 2.0. With a maximum possible score of 210 points, the new AMD Radeon HD 6870 graphics card scores 198 points, while the top competitor scores only 138 points. However, this is a test by AMD itself, and such results should always be treated with caution. Not because of deceit, but often craftiness.

It seems to us that a very interesting novelty is the appearance of support for decoding the DivX / XviD format (read, MPEG-4). But not only this format has received improvements, now MPEG-2 is fully decoded on the GPU, and AMD also has support for codecs with two streams (Blu-ray 3D).

And yet, it is more interesting that the newly released video cards from AMD, thanks to the inclusion of the latest modification of the third generation UVD block in the GPU, are able to speed up the playback of MPEG-4 videos. This is important not only and not so much because of the reduced CPU usage during decoding itself, but will help extend the battery life of laptops and netbooks, reduce the noise from the fans of PC-based home theaters (HTPC) and allow you to play high-resolution MPEG-4 files. on budget PCs.

At the event for journalists, a demonstration of simultaneous decoding on the CPU and GPU was shown. As you can see, with fully software decoding, the CPU is loaded with work by more than 20%, and when shifting work to the AMD GPU, the central processor of the system practically ceases to perform any significant work, because it becomes 10 times smaller. It is clear that all this was done before, but not for the DivX/XviD format.

Non-graphic calculations

In this sense, there are no hardware changes in Barts, but they are in the software part. AMD prefers to refer to GPU computing as Parallel Processing. And of course, they only support industrial standards - open OpenCL and closed, but no less industrial DirectCompute from DirectX 11.

OpenCL attracts AMD as an open and multi-platform API for the so-called heterogeneous architectures, which is very well suited for the same AMD Fusion. It is with the help of OpenCL that you can unlock the computing capabilities of both the CPU and the GPU. It is clear that AMD was the first company to introduce OpenCL for CPU and GPU at the same time. But in general, OpenCL is supported by such large companies as Apple, IBM, Intel, NVIDIA, Sony, etc.

DirectCompute has other advantages: being distributed as part of DirectX by Microsoft, and a very simple method for incorporating GPU computing into existing DirectX applications, and especially 3D games.

Changes in AMD parallel computing have come more in terms of names than hardware. The ATI Stream brand has been replaced by AMD Accelerated Parallel Processing (APP) technology. It's a bit long in my opinion, although it better describes what the technology means and is quite in line with the widespread abandonment of the ATI brand. The company decided to make changes in the brand right now, with the announcement of a new generation of graphics cards and the release of a new line, which is absolutely logical.

The SDK is now called AMD APP SDK (formerly ATI Stream SDK) and includes a complete OpenCL development platform for GPUs and multi-core x86 CPUs, and AMD Fusion is also supported. The company's website now has an OpenCL Zone section that looks suspiciously like the CUDA Zone, where developers can find up-to-date information on OpenCL, tutorials on working with OpenCL, developer tools and various libraries, and any other material on the topic.

Details: Antilles, Radeon HD 6990 series

• Codename "Antilles"
• 40 nm technology
• 2 chips with 2.64 billion transistors each
• The area of ​​each crystal is 389 mm2
• Unified architecture with an array of common processors for streaming processing of multiple types of data: vertices, pixels, and more.
• Hardware support for DirectX 11, including a new shader model - Shader Model 5.0
• Dual 256-bit memory bus: twice four 64-bit wide controllers with GDDR5 memory support
• Core clock 830 to 880 MHz (see explanation below)
• 2x24 SIMD cores, including 768 stream processors, and a total of 3072 scalar floating point ALUs (integer and float formats, support for FP32 and FP64 precision within the IEEE 754 standard)
• 2x24 large texture units, with support for FP16 and FP32 formats
• 2x96 ​​texture address units and the same number of bilinear filtering units, with the ability to filter FP16 textures at full speed and support for trilinear and anisotropic filtering for all texture formats
• 2x32 ROPs with support for anti-aliasing modes with the possibility of programmable sampling of more than 16 samples per pixel, including with FP16 or FP32 framebuffer format. Peak performance up to 64 samples per clock (including for FP16 buffers), and in colorless mode (Z only) - 256 samples per clock
• For each GPU, integrated support for RAMDAC, six Single Link or three Dual Link DVI ports, plus HDMI 1.4a and DisplayPort 1.2

Radeon HD 6990 (HD 6990 OC) Graphics Card Specifications

• Core clock 830(880) MHz
• Number of universal processors 3072
• Number of texture blocks - 2x96, blending blocks - 2x32
• Effective memory frequency 5000 MHz (4×1250 MHz)
• Memory type GDDR5
• Memory capacity 2x2 gigabytes
• Memory bandwidth 2x160 gigabytes per second.
• The theoretical maximum fill rate is 53 (56) gigapixels per second.
• Theoretical texture sampling rate is 159 (169) gigatexels per second.
• Crossfire connector
• PCI Express 2.1 bus
• Connectors: DVI Dual Link, four mini DisplayPort 1.2
• Power consumption from 37 to 375(450) W
• Typical gaming power consumption - up to 350(415) W
• Two 8-pin power connectors
• Two-slot execution;
• The recommended price for Russia is 22999 rubles. (for the USA - $699).

As we mentioned earlier, in this generation of AMD video cards, the principle of naming models has been changed. Since the HD 5870 and HD 5850 video cards were replaced by two lines at once: HD 6800 and HD 6900, and the latter received the fastest GPU, it is quite logical that a dual-chip card based on the same GPUs also entered the HD 6900 series. But since the index 6970 was already occupied by a top-end single-chip solution, so the new video card got the 6990 index. That is, compared to the previous similar HD 5970 board, not only the first, but also the third digit of the index has changed.

The new AMD graphics card is equipped with GDDR5 memory and 2 gigabytes of memory per GPU. This decision is quite justified for a product of this level, because in some gaming applications with maximum settings, high resolution and maximum level anti-aliasing enabled, 1 gigabyte of memory per chip is no longer enough today. And even more so when rendering in stereo or on three monitors in Eyefinity mode at ultra-high resolutions.

Naturally, the video card has a two-slot cooling system, which is quite long and covered with a plastic casing, familiar to all modern AMD motherboards, along the entire length. The power consumption of a card with two GPUs on board is quite high for obvious reasons, so we had to install two 8-pin power connectors on it, which was not previously seen in reference samples (although some video card manufacturers made such solutions on their own).

Architecture

Since the Antilles video card is based on two GPUs of the Cayman family, there is simply no point in talking about this section especially - everything has already been done earlier, in the corresponding article. However, let's recap the basics. The goal of AMD engineers was to create an efficient graphics and computing architecture with enhanced GPGPU capabilities, as well as the implementation of geometric block parallelism and improvements in texture filtering and full-screen anti-aliasing.

The Cayman architecture became an intermediate solution between the previous Cypress architecture and the unborn 32 nm architecture, which was not destined to enter the market. But the composition of the new GPU still included some of the features from it. The extra transistors compared to the Cypress were spent on new computing and graphics capabilities.

The most important thing in the GPU is the two graphics engines, including the rasterizer, tesselator and other geometry processing units, as well as a dual dispatcher. The dual geometry block in AMD's top GPU is now able to process two primitives per clock, i.e., the speed of transformation and discarding the back faces has doubled, and together with improved buffering - up to three times in some cases, compared to Cypress-based solutions.

Another major architectural change was the superscalar VLIW4 architecture of computing processors, unlike the previous VLIW5. Each stream processor has 4 ALUs instead of 5 as was previously the case. This decision increased the efficiency of using stream processors, although it reduced the potential peak performance. For more information on the Cayman architecture, see the baseline review linked above.

Power and cooling

When designing video cards with two of the most powerful GPUs on the same board and their serious power requirements, maximum attention should be riveted to the corresponding system. That's why the new generation digital programmable voltage regulators manufactured by Volterra, as well as powerful four-phase power inductors manufactured by Cooper Bussmann of the CL1108 series, are used in the power supply circuit of the Radeon HD 6990.

All this resulted in an increase in the efficiency of the power circuit, compared to previous devices used by AMD, and therefore lower temperatures and lower power consumption. In addition, the symmetrical layout of the regulators in the center of the printed circuit board also worked to increase efficiency.

Efficient cooling of such a hot two-chip solution is perhaps an even more important and difficult task. The Radeon HD 6990 cooler uses a new preinstalled phase-change thermal interface. It is recognized by AMD as 8% more efficient than previous materials used for this task. The figure may seem small, but in the matter of cooling such extreme devices, every little thing counts.

The new cooler itself uses two vapor chambers (one for each GPU) and a single fan located between them in the center of the board. It handles up to 450W of heat in and out quite well, and although the new board is exactly the same size as the Radeon HD 5970, all of the above improvements result in the new cooler being noticeably more efficient than the previous solution.

AMD PowerTune technology

Support for this technology on the dual-chip video card Radeon HD 6990 is expected. It is in the case of such power-demanding boards that it is imperative to control the power consumption and limit it if something happens. The technology was first announced with the Radeon HD 6970 and HD 6950, and in the basic article about them, we described its operation in as much detail as possible. Therefore, we will repeat only the most important points.

The Cayman series GPUs have special sensors in the execution units that monitor the load parameters, and the GPU constantly monitors the load and power consumption, and does not allow the latter to go beyond a certain threshold, automatically changing the frequency and voltage so that these parameters remain within a certain heat pack. The technology helps to set relatively high GPU frequencies and at the same time not be afraid of a video card failure due to exceeding safe power consumption limits.

The technology is useful for several reasons. It protects video cards from failure in case of inadequate overclocking experiments, and also allows you to squeeze maximum performance out of the GPU. In addition, PowerTune allows the user to change the consumption limit himself using AMD OverDrive tools within certain limits (plus or minus 20%). Naturally, adjusting the maximum consumption parameter deprives the user of any guarantees.

It is important that PowerTune technology is aimed at obtaining maximum performance in gaming applications, and not stability tests, which often inadequately heavily load all GPU units at once. As you can see in the diagram above, the technology allows you to increase the GPU clock speeds in games, maintaining the set level of power consumption and not requiring software solutions in the video driver code, as is done in a similar (but much simplified) competitor's technology.

BIOS switch (Dual-BIOS)

When the Radeon HD 6970 and HD 6950 had a switch between two BIOS versions, it immediately became clear that this was not only and not so much a solution aimed at greater reliability, but a solution that allowed bold experiments on a video card. Moreover, not only for users, but also for video card manufacturers. Actually, this is what happened - some of the manufacturers recorded not just a version with factory increased frequencies as a second BIOS image, but even an image from an older video card model, turning the Radeon HD 6950 into an HD 6970.

It is logical that a similar solution appeared in the Radeon HD 6990. Moreover, it even received further development. The switch between the two BIOS versions in the new solution, even in the reference version, allows you to enable super mode (uber mode) - with increased GPU clock frequencies from 830 MHz to 880 MHz and voltage from nominal 1.12 V to 1.175 V. Naturally, the amount of energy consumed also increases significantly at the same time , and most likely it was for this mode that two 8-pin additional power connectors were installed on the board.

Switch position "2" is the nominal mode with a frequency of 830 MHz, in this position the video card is delivered. BIOS switch mode "1" enables factory overclocking and is intended for overclockers and enthusiasts who understand that this mode will require a significantly more powerful power supply and improved cooling in the case.

Attention! Despite the fact that factory overclocking is now enabled on absolutely all Radeon HD 6990s using the BIOS switch, this does not mean at all that the company takes on warranty obligations in the event of a video card failure due to overclocking! AMD's warranty does not cover such cases, no matter how the graphics card was overclocked, using software driver settings in the Catalyst Control Center or using the Dual-BIOS switch.

Apparently, AMD is aware that video cards like the Radeon HD 6990 are bought only by enthusiasts and overclockers, who for the most part know how to prevent video card failure with a small (880 MHz) overclock, but just in case, it protects itself from extreme would-be overclockers who they burn video cards like a forgetful granny with her pies in the oven.

Although even for ordinary users there is a sense in such a pre-overclocked mode - an extra 5-6% (in reality, most often about 3-4%) to performance will not interfere if the PSU is good and the cooling in the case is arranged correctly. After all, for automatic overclocking, now you just need to move the switch lever, and everything else is already done.

AMD Eyefinity Technology

This multi-monitor technology from AMD has long been known to our readers. In fact, all the company's video cards support Eyefinity, the best multi-monitor system at the moment, supporting up to six monitors even in the case of single-chip solutions. The only thing is that support for six monitors at the same time will require the use of special hubs that are compatible with multi-stream signal transmission over DisplayPort - Multi-Stream Transport.

But even without the use of hubs, any of the two dozen currently produced AMD Radeon models support connecting three monitors in various configurations. And to support Eyefinity, games only need to be able to work with non-standard resolutions and aspect ratios. At the moment, about 70 games can boast of proven support for the technology, and hundreds more applications are compatible with it.

Moreover, it is such a powerful solution as the Radeon HD 6990 that will allow you to comfortably play on three monitors with a total resolution of 7680x1600 or five vertical ones with a resolution of 6000x1920, delivering 30 frames per second or more even in heavy games, which was previously unavailable for single video cards. Although such modes remain more the lot of exhibitions and various events than ordinary home users, who would rather prefer a projector or a huge TV instead of five monitors on a poor table.

Due to the need for effective cooling, and in particular - the maximum removal of heated air, the set of video signal outputs had to be changed. Exactly half of the area of ​​the slot plug was occupied by the exhaust holes of the cooling system. And on the remaining part they placed one Dual Link DVI connector and four mini DisplayPort 1.2 connectors. Thus, with all the limitations of a powerful cooler, we managed to keep the maximum possible number of pins.

But after all, for this you need to look for quite rare and not so cheap adapters with mini DisplayPort, the caustic reader will ask? Not at all necessary. Each Radeon HD 6990 graphics card will come with a set of three such adapters: passive mini DisplayPort - Single Link DVI, active mini DisplayPort - Single Link DVI and passive mini DisplayPort - HDMI.

Details: Barts LE, Radeon HD 6700 series

• Chip codename "Barts"
• 40 nm technology
• 1.7 billion transistors
• Unified architecture with an array of common processors for streaming processing of multiple types of data: vertices, pixels, and more.
• Hardware support for DirectX 11, including a new shader model - Shader Model 5.0
• 256-bit memory bus: four 64-bit wide controllers with GDDR5 memory support
• Core clock up to 840 MHz
• 14 (10 active) SIMD cores, including 1120 (800 active) floating point scalar ALUs (integer and float formats, supports IEEE 754 FP32 precision)
• 14 (10 active) enlarged texture units, with support for FP16 and FP32 formats
• 56 (40 active) texture address units and the same number of bilinear filtering units, with the ability to filter FP16 textures at full speed and support for trilinear and anisotropic filtering for all texture formats
• 32 (16 active) ROPs with support for anti-aliasing modes with the possibility of programmable sampling of more than 16 samples per pixel, including with FP16 or FP32 framebuffer format. Peak performance up to 16 samples per clock (including for FP16 buffers), and in colorless mode (Z only) - 64 samples per clock
• Write results to eight frame buffers simultaneously (MRT)
• Integrated support for RAMDAC, six Single Link or three Dual Link DVI ports, plus HDMI 1.4a and DisplayPort 1.2

Radeon HD 6790 Specifications

• Core clock 840 MHz
• Number of universal processors 800
• Number of texture blocks - 40, blending blocks - 16
• Effective memory frequency 4200 MHz (4×1050 MHz)
• Memory type GDDR5
• Memory size 1024 megabytes
• Memory bandwidth 134.4 gigabytes per second.
• Theoretical maximum fill rate is 13.4 gigapixels per second.
• Theoretical texture fetch rate is 33.6 gigatexels per second.
• CrossFireX support
• PCI Express 2.1 bus
• Connectors: DVI Dual Link, DVI Single Link, HDMI 1.4a, two mini DisplayPort 1.2
• Power consumption 19 to 150 W (two 6-pin power connectors)
• Dual slot design
• MSRP for the US market $149

The use of the same Barts chip in a solution of this level became possible due to the improved characteristics of the 40 nm process technology, as well as the desire to get rid of rejected chips. Unfortunately, the new solution cannot be called particularly energy efficient, since its maximum consumption level is set even higher than that of the same Radeon HD 6850. Apparently, this was done in order to increase the voltage on the GPU along with the clock frequency, and at the same time use a larger part of the chips that previously went to the wastebasket.

The new AMD video card will have to compete with solutions based on the NVIDIA GeForce GTX 550 Ti, which came out quite a lot, including overclocked ones, and with different amounts of video memory. You will also have to fight with options like the GeForce GTX 460, which have been on sale for a long time and managed to get very cheap, so when choosing a video card in this price range, they will also be drawn to the attention of a potential buyer.

The principle of naming models remains the same as in the latest solutions of the company. Compared to other solutions, not only the second, but also the third digit in the index has changed. For some strange reason, it suddenly became not 7, as was previously accepted (5870, 6870, 6970), but 9. Apparently, this should indicate a very small difference in performance between the Radeon HD 6850 and HD 6790.

It is quite logical that one gigabyte of GDDR5 memory is installed on the video card. This is the optimal amount of memory today, even for solutions from the lower price range. Interestingly, although the width of the video memory bus in the HD 6790 remained 256-bit, the number of ROPs was halved, from 32 to 16. We have already seen such a solution in previous "truncated" AMD products.

Despite belonging to the lower price range, the new video card has a two-slot cooling system, covered by a plastic casing already familiar to AMD cards along the entire length (however, we are talking about a reference design, and manufacturers will most often make their own motherboards and coolers). We have already talked about energy consumption, it is quite high. That is why I had to install not one, but two whole 6-pin auxiliary power connectors.

Architecture

We have already covered the Barts GPU architecture in the corresponding background article, and you should refer to it for all the details. As you remember, this chip is a development of the ideas of previous generations, and the differences between Barts and Cypress are mostly quantitative, although not only.

As in the case of the latest competitor GPUs, Barts basically improved the performance per watt and millimeter of area consumed, that is, improved efficiency, compared to previous GPUs. But still, Barts cannot be called a completely new chip, because compared to the previous ones, it simply has a different number of execution units and a changed balance between performance and consumption.

Small optimizations have led to an increase in the speed of geometry processing, but this has not changed the situation particularly noticeably, in tessellation problems, the competitor's solutions remain stronger. More interesting is the support of new video chips with UVD3 for decoding video data of DivX formats, as well as Blu-ray 3D video, and improvements in AMD Eyefinity and DisplayPort 1.2 support.

What has changed in the GPU compared to the Radeon HD 6870 and HD 6850? In fact, some of the 14 available hardware SIMD blocks, as well as half of the ROP blocks, are simply disabled in the video chip. Accordingly, the total number of stream processing processors has also decreased, now there are only 800 of them, unlike 1120 for a full-fledged Barts. But the ROP blocks were not 32 at all, but only 16. Everything else remained the same, even the 256-bit memory bus.

Due to fairly high clock speeds and not too much cut down on the main execution units of the GPU (the fillrate may be lacking only in rare cases and with anti-aliasing turned on, most likely), the performance of the Radeon HD 6790 should be almost the same as that of the HD 6850, and at the same time, it is slightly higher than that of the HD 5770. And at the same time, the new Radeon model should outperform the main rival in the face of the GeForce GTX 550 Ti.

Nvidia. This review discusses the pros and cons of the AMD Radeon HD 6800 Series. Characteristics, description and test results - all this you can find below.

The advent of a series of video cards

AMD regularly updates the line of graphics processors and video cards. 2010 was no exception: the 6800 series was introduced to the public. This line was created to replace the flagship 5870 video card.

On October 22, the video card was presented Feedback on the course of the presentation of the line was only positive. In 2010, AMD was only gaining popularity with its video cards, so everyone was expecting a technical breakthrough from them, or at least a very good flagship series.

It was on this line that the manufacturer's rebranding completely ended: from now on, to this day, video cards were called AMD, not ATI. This was done due to the termination of the contract after the merger of the companies. Perhaps this decision was made to popularize not only graphics chips, but also processors from AMD. The conclusion about this suggests itself due to constant advertising and the presentation of configurations assembled only on the AMD platform (processor + video card).

Let's figure out what the AMD Radeon HD 6800 Series line has brought to the market of video cards for desktop computers, the characteristics of which will be presented below. The entire series is represented by the following video cards: HD 6850 and 6870. According to the creators themselves, the number 8 in the index no longer means belonging to the top line of graphics chips, since the 6900 series has appeared.

AMD Radeon HD 6800 Series Specifications

First, it is worth talking about changing the platform. The new line uses the Barts processor. From the first presentation, it became clear that AMD chose a different development path than Nvidia. If the latter are constantly in pursuit of power and maximum performance, then Radeon video cards are designed to be a balanced ratio, no matter how trite it may sound, price and quality (performance).

Specialists from the former ATI company were often called real innovators. They set trends for the entire graphics chip market. After moving under the wing of AMD, the company took a step back. The new generation of Barts processors is even weaker than the previous one on paper and in specifications. The creators went by simplifying the architecture to achieve an excellent balance between speed, reliability and performance. Barts has become simpler in structure and smaller in size. This processor is the basis for the middle class, which includes the AMD Radeon HD 6800 Series. Specifications are shown below.

Both representatives of the series and 6870) support DirectX11 and version 5 shaders. The cost of video cards is 180 and 240 dollars, respectively. Compared to Nvidia's fast and overclocked competitors, AMD's motherboards are truly budget-friendly, but the difference in performance isn't that great. on both cards - 1 GB. The series is a direct competitor to the GeForce GTX460 with 1 GB of RAM and the GeForce GTX470.

AMD Radeon HD 6800 Series graphics card: specifications and test results

To test the line of video cards, the following was used as a test bench: a Core i7 processor with a frequency of 3.3 GHz, 6 GB of RAM and a 64-bit Windows 7 operating system. All games used are set to graphics quality and detail to test the maximum performance of the tested video cards.

The first test game was Aliens vs. Predator. It immediately becomes clear that the HD6800 series will be hard to compete with the GeForce 460 1GB: only at a resolution of 1600x900 and lower can an AMD card produce playable 30 frames per second.

In the game Battlefield Bad Company 2, the situation is evened out, and it does not seem like such a bad decision to purchase an AMD Radeon HD 6800 Series. Specifications at maximum graphics and resolution settings (6850 and 6870) allow you to overtake GeForce by as much as 8 frames per second (30 versus 22). Recall that the cost of an Nvidia graphics card is from $230. The use of the new line from AMD is becoming more and more attractive. But without doing look at the following tests.

In the very demanding game Crysis Warhead, both video cards hold up decently only at low screen resolutions. STALKER Call of Pripyat gives Nvidia's graphics card a 10fps lead. But do not forget about the significant difference in price.

Conclusion after tests

In general, the AMD Radeon HD 6800 Series graphics card shows itself worthy in all games. After the update, the drivers began to support all new games, so the budget version of the AMD graphics chip produces a tolerable 25-30 frames per second in modern game projects at high graphics settings.

AMD Radeon HD 6800 Series: pros and cons

The following points can be distinguished from the advantages of this video card. Firstly, good performance in most modern games. Second, low power consumption. You can also note the low cost, for which the buyer will receive good performance and all the "chips" of top-end video cards, such as displaying an image on 6 monitors, compatibility mode with similar video cards.

The disadvantages are hidden in the increased noise of the video card and a frankly weak cooling system. At sufficiently high loads in video games, the chip starts to overheat quickly.

Outcome

For those who are not looking for breakthrough power and high numbers in tests, the AMD Radeon HD 6800 Series line is perfect. The characteristics of video cards allow you to safely play with high FPS at medium or close to them settings for the graphic component of the game. On the side of video cards from AMD, there is also a low cost compared to Nvidia GeForce 460 and 470. But the performance differs little, so the choice of a mid-range budget video card is obvious.

Positioning among AMD video cards

AMD's slide illustrates the placement of the AMD Radeon HD 6800 series in the AMD product hierarchy:

As you can see, some reforms have been outlined here. Two new solutions have been prepared to replace the ATI Radeon HD 5800 series video cards:
AMD Radeon HD 6800 graphics cards will introduce AMD products in the $150-250 price range, and in the near future, solutions based on AMD "Cayman" GPUs will take a step up. ATI Radeon HD 5700 series video cards will continue to exist for the time being.

The next slide demonstrates the positioning of new video card lines in relation to performance:

Thus, at the beginning of 2011, instead of three lines of AMD video cards, the market will be occupied by four lines. In the fourth quarter of 2010, AMD Radeon HD 6990 accelerators will be released to replace the ATI Radeon HD 5970, leading the highest level of AMD products. The AMD Radeon HD 6950 and Radeon HD 6970 graphics cards will be placed below, while the performance of the Radeon HD 6900 should significantly exceed the current line of ATI Radeon HD 5850 and Radeon HD 5870 accelerators. The new AMD Radeon HD 6800 actually replaces the ATI Radeon HD 5800 series line. ATI Radeon HD 5770 will remain on the market for the time being, closing the AMD lineup.

As a result of the reforms, we get AMD Radeon HD 6800 video cards at a price of $150-250, with performance almost comparable to more expensive ATI Radeon HD 5800 adapters.

Market positioning

AMD Radeon HD 6800 should compete with NVIDIA GeForce GTX 460 graphics cards. In the future, NVIDIA GeForce GTX 470 and GeForce GTX 480 accelerators will be opposed by solutions based on AMD's "Cayman" graphics processors, which have not yet been introduced. AMD Radeon HD 6990 ("Antilles") video cards should lead the market, but this is in the near future, but for now let's return to the already announced AMD Radeon HD 6800.

Together with the announcement of the AMD Radeon HD 6800 series video cards, NVIDIA announced a reduction in the recommended cost for some accelerators.

As a result, the market for graphics accelerators in the range of $150-250 looks like this:

Nvidia. This review discusses the pros and cons of the AMD Radeon HD 6800 Series. Characteristics, description and test results - all this you can find below.

The advent of a series of video cards

AMD regularly updates the line of graphics processors and video cards. 2010 was no exception: the 6800 series was introduced to the public. This line was created to replace the flagship 5870 video card.

On October 22, the AMD Radeon HD 6800 Series video card was presented. Feedback on the course of the presentation of the line was only positive. In 2010, AMD was only gaining popularity with its video cards, so everyone was expecting a technical breakthrough from them, or at least a very good flagship series.

It was on this line that the manufacturer's rebranding completely ended: from now on, to this day, video cards were called AMD, not ATI. This was done due to the termination of the contract after the merger of the companies. Perhaps this decision was made to popularize not only graphics chips, but also processors from AMD. The conclusion about this suggests itself due to constant advertising and the presentation of configurations assembled only on the AMD platform (processor + video card).

Let's figure out what the AMD Radeon HD 6800 Series line has brought to the market of video cards for desktop computers, the characteristics of which will be presented below. The entire series is represented by the following video cards: HD 6850 and 6870. According to the creators themselves, the number 8 in the index no longer means belonging to the top line of graphics chips, since the 6900 series has appeared.

AMD Radeon HD 6800 Series Specifications

First, it is worth talking about changing the platform. The new line uses the Barts processor. From the first presentation, it became clear that AMD chose a different development path than Nvidia. If the latter are constantly in pursuit of power and maximum performance, then Radeon video cards are designed to be a balanced ratio, no matter how trite it may sound, price and quality (performance).

Specialists from the former ATI company were often called real innovators. They set trends for the entire graphics chip market. After moving under the wing of AMD, the company took a step back. The new generation of Barts processors is even weaker than the previous one on paper and in specifications. The creators went by simplifying the architecture to achieve an excellent balance between speed, reliability and performance. Barts has become simpler in structure and smaller in size. This processor is the basis for the middle class and budget video cards, which include the AMD Radeon HD 6800 Series. Specifications are shown below.

Both representatives of the series (HD 6850 and 6870) support DirectX11 and version 5 shaders. The cost of video cards is 180 and 240 dollars, respectively. Compared to Nvidia's fast and overclocked competitors, AMD's motherboards are truly budget-friendly, but the difference in performance isn't that great. The amount of video memory on both cards is 1 GB. The series is a direct competitor to the GeForce GTX460 with 1 GB of RAM and the GeForce GTX470.

AMD Radeon HD 6800 Series graphics card: specifications and test results

To test the line of video cards, the following computer configuration was used as a test bench: a Core i7 processor with a frequency of 3.3 GHz, 6 GB of RAM and a 64-bit Windows 7 operating system. All games used are set to graphics quality and detail to test the maximum performance of the tested video cards.

The first test game was Aliens vs. Predator. It immediately becomes clear that the HD6800 series will be hard to compete with the GeForce 460 1GB: only at a resolution of 1600x900 and lower can an AMD card produce playable 30 frames per second.

In the game Battlefield Bad Company 2, the situation is evened out, and it does not seem like such a bad decision to purchase an AMD Radeon HD 6800 Series. Specifications at maximum graphics and resolution settings (6850 and 6870) allow you to overtake GeForce by as much as 8 frames per second (30 versus 22). Recall that the cost of an Nvidia graphics card is from $230. The use of the new line from AMD is becoming more and more attractive. But without jumping to conclusions, let's look at the following tests.

In the very demanding game Crysis Warhead, both video cards hold up decently only at low screen resolutions. STALKER Call of Pripyat gives Nvidia's graphics card a 10fps lead. But do not forget about the significant difference in price.

Conclusion after tests

In general, the AMD Radeon HD 6800 Series graphics card shows itself worthy in all games. After the update, the drivers began to support all new games, so the budget version of the AMD graphics chip produces a tolerable 25-30 frames per second in modern game projects at high graphics settings.

AMD Radeon HD 6800 Series: pros and cons

The following points can be distinguished from the advantages of this video card. Firstly, good performance in most modern games. Second, low power consumption. You can also note the low cost, for which the buyer will receive good performance and all the "chips" of top-end video cards, such as displaying an image on 6 monitors, compatibility mode with similar video cards.

The disadvantages are hidden in the increased noise of the video card and a frankly weak cooling system. At sufficiently high loads in video games, the chip starts to overheat quickly.

Outcome

For those who are not looking for breakthrough power and high numbers in tests, the AMD Radeon HD 6800 Series line is perfect. The characteristics of video cards allow you to safely play with high FPS at medium or close to them settings for the graphic component of the game. On the side of video cards from AMD, there is also a low cost compared to Nvidia GeForce 460 and 470. But the performance differs little, so the choice of a mid-range budget video card is obvious.

AMD regularly updates the line of graphics processors and video cards. 2010 was no exception: the 6800 series was introduced to the public. This line was created to replace the flagship 5870 video card.

On October 22, the AMD Radeon HD 6800 Series video card was presented. Feedback on the course of the presentation of the line was only positive. In 2010, AMD was only gaining popularity with its video cards, so everyone was expecting a technical breakthrough from them, or at least a very good flagship series.
It was on this line that the manufacturer's rebranding completely ended: from now on, to this day, video cards were called AMD, not ATI. This was done due to the termination of the contract after the merger of the companies. Perhaps this decision was made to popularize not only graphics chips, but also processors from AMD. The conclusion about this suggests itself due to constant advertising and the presentation of configurations assembled only on the AMD platform (processor + video card).


Let's figure out what the AMD Radeon HD 6800 Series line has brought to the market of video cards for desktop computers, the characteristics of which will be presented below. The entire series is represented by the following video cards: HD 6850 and 6870. According to the creators themselves, the number 8 in the index no longer means belonging to the top line of graphics chips, since the 6900 series has appeared.

AMD Radeon HD 6800 Series Specifications

First, it is worth talking about changing the platform. The new line uses the Barts processor. From the first presentation, it became clear that AMD chose a different development path than Nvidia. If the latter are constantly in pursuit of power and maximum performance, then Radeon video cards are designed to be a balanced ratio, no matter how trite it may sound, price and quality (performance).
Specialists from the former ATI company were often called real innovators. They set trends for the entire graphics chip market. After moving under the wing of AMD, the company took a step back. The new generation of Barts processors is even weaker than the previous one on paper and in specifications. The creators went by simplifying the architecture to achieve an excellent balance between speed, reliability and performance. Barts has become simpler in structure and smaller in size. This processor is the basis for the middle class and budget video cards, which include the AMD Radeon HD 6800 Series. Specifications are shown below.


Both representatives of the series (HD 6850 and 6870) support DirectX11 and version 5 shaders. The cost of video cards is 180 and 240 dollars, respectively. Compared to Nvidia's fast and overclocked competitors, AMD's motherboards are truly budget-friendly, but the difference in performance isn't that great. The amount of video memory on both cards is 1 GB. The series is a direct competitor to the GeForce GTX460 with 1 GB of RAM and the GeForce GTX470.

AMD Radeon HD 6800 Series graphics card: specifications and test results

To test the line of video cards, the following computer configuration was used as a test bench: a Core i7 processor with a frequency of 3.3 GHz, 6 GB of RAM and a 64-bit Windows 7 operating system. All games used are set to graphics quality and detail to test the maximum performance of the tested video cards.
The first test game was Aliens vs. Predator. It immediately becomes clear that the HD6800 series will be hard to compete with the GeForce 460 1GB: only at a resolution of 1600×900 and lower can an AMD card produce playable 30 frames per second.


In the game Battlefield Bad Company 2, the situation is evened out, and it does not seem like such a bad decision to purchase an AMD Radeon HD 6800 Series. Specifications at maximum graphics and resolution settings (6850 and 6870) allow you to overtake GeForce by as much as 8 frames per second (30 versus 22). Recall that the cost of an Nvidia graphics card is from $230. The use of the new line from AMD is becoming more and more attractive. But without jumping to conclusions, let's look at the following tests.
In the very demanding game Crysis Warhead, both video cards hold up decently only at low screen resolutions. STALKER Call of Pripyat gives Nvidia's graphics card a 10fps lead. But do not forget about the significant difference in price.

AMD Radeon HD 6800 Series: pros and cons

The following points can be distinguished from the advantages of this video card. Firstly, good performance in most modern games. Second, low power consumption. You can also note the low cost, for which the buyer will receive good performance and all the "chips" of top-end video cards, such as displaying an image on 6 monitors, compatibility mode with similar video cards.


The disadvantages are hidden in the increased noise of the video card and a frankly weak cooling system. At sufficiently high loads in video games, the chip starts to overheat quickly.

Outcome

For those who are not looking for breakthrough power and high numbers in tests, the AMD Radeon HD 6800 Series line is perfect. The characteristics of video cards allow you to safely play with high FPS at medium or close to them settings for the graphic component of the game. On the side of video cards from AMD, there is also a low cost compared to Nvidia GeForce 460 and 470. But the performance differs little, so the choice of a mid-range budget video card is obvious.