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

The emergence of a series of video cards

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

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

It was on this line that the rebranding of the manufacturer came to an end: from now on, video cards were named by AMD, not ATI. This was done due to the expiration 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 the constant advertising and presentation of configurations collected only on the AMD platform (processor + video card).

Let's figure out what's new brought to the market of video cards for stationary computers by the AMD Radeon HD 6800 Series, 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 appeared.

AMD Radeon HD 6800 Series Specifications

First, it's worth talking about the platform change. The new line uses the Barts processor. From the first presentation, it became clear that AMD chose a different path of development than Nvidia. If the latter are constantly in pursuit of power and maximum performance, then Radeon video cards are created to be a balanced ratio, no matter how corny it sounds, 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 mid-range and budget graphics cards, which include the AMD Radeon HD 6800 Series. The characteristics are presented below.

Both representatives of the series (HD 6850 and 6870) support DirectX11 and shader version 5. The cost of video cards is 180 and 240 dollars, respectively. Compared to the performance and overclocked competitors from Nvidia, the boards from AMD are truly budget-friendly, but the difference in performance is not so great. The video memory on both cards is 1 GB. The series is a direct competitor to the GeForce GTX460 with 1GB of RAM and the GeForce GTX470.

AMD Radeon HD 6800 Series Graphics Card Specifications and Benchmarks

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 tuned to the quality of graphics and detail to check the maximum performance of the tested video cards.

The first game of the test was Aliens vs. Predator. It immediately becomes clear that the HD6800 series will be difficult to compete with the GeForce 460 1GB: only at 1600x900 and below the AMD motherboard can deliver playable 30 frames per second.

In the game Battlefield Bad Company 2, the situation is leveled, and the purchase of AMD Radeon HD 6800 Series does not seem to be such a bad decision. Technical characteristics at the maximum graphics and resolution settings (6850 and 6870) allow you to overtake the GeForce by as much as 8 frames per second (30 versus 22). Recall that the cost of an Nvidia video card starts at $ 230. Using 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 well only at low screen resolutions. STALKER Call of Pripyat gives Nvidia's graphics card a 10fps lead. But don't forget about the significant price difference.

Conclusion after tests

In general, in all games, the AMD Radeon HD 6800 Series video card shows itself worthy. After the update, the drivers began to support all new games, so the budget version of the graphics chip from AMD 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. First, good performance in most modern games. Secondly, low power consumption. You can also note the low cost, for which the buyer will receive good performance and all the "chips" of top video cards, such as displaying images on 6 monitors, compatibility mode with similar video cards.

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

Outcome

For those who are not chasing breakthrough capacities and high numbers in tests, the AMD Radeon HD 6800 Series line is perfect. The characteristics of the video cards allow you to play safely with high FPS at medium or close to them settings of the graphics component of the game. On the side of AMD video cards, the price is also low, compared to the Nvidia GeForce 460 and 470. But the performance does not differ much, so the choice of a video card in the middle budget class is obvious.

The ATI Radeon HD 5800 video cards, released last fall, had a strong impact on the market, securing AMD as the supplier of the fastest solutions for gaming PCs for almost six months. The release of NVIDIA Fermi in the spring of 2010 forced the company to make room for a little, and now AMD is once again playing a trump card, playing on the price-performance ratio.

Reference AMD Radeon HD 6850

HIS Radeon HD 6870

The ATI Cypress graphics architecture underlying the Radeon HD 5000 was undoubtedly a new milestone in the evolution of the Canadian developer's video cards: the performance gain compared to the previous generation was so significant that the company triumphantly returned with top-end models to the “over $ 300” price niche. However, GPU developers do not receive their main income from high-end products, which are the prerogative of enthusiasts, but from mid-range video cards available to the mass buyer. It was this niche that paid the most attention to AMD, developing a new generation, codenamed Northern Islands. Its first representatives were AMD Radeon HD 6870 and HD 6850, also referred to as "Barts" (note that these accelerators were the first to bear the name of AMD, not ATI - the corporation is trying to strengthen consumers' perception of its brand as a representative of a solid platform with processors, chipsets and GPU).

Note that the priority on video cards for the price range "up to $ 250" does not mean that AMD does not plan to upgrade the top segment: the most productive Radeon HD 6900 products with one processor (Cayman) and two (Antilles) will be released later, but for now the company is concentrating on a problematic niche for yourself. The fact is that the Radeon HD 5870 and HD 5850 are based on a rather complex Cypress GPU, which is expensive to manufacture, and as a result of the price confrontation with NVIDIA, it no longer brings enough profit to AMD. Positioned below the Radeon HD 5770, on the other hand, are profitable for the company, but not very attractive to the consumer: the GeForce GTX 460 has a very high performance at an affordable price. AMD solves this problem by replacing the older single-chip Radeon HD 5800 models with a new generation with a reduced price , sharpening competition in the mainstream segment. NVIDIA, in turn, has already made certain sacrifices in order to remain competitive: the cost of the GeForce GTX 460 with 1 GB of memory, which is the main competitor of AMD's new products, has been reduced from $ 229 to $ 199, and the most surprising is the price of the GeForce GTX 470 with the expensive top-end GPU GF100 also dropped from $ 349 to $ 259. And this vendor has serious reasons for concern.

The architecture of AMD Barts clearly traces the legacy of Cypress, these GPUs inherited the structure of computational and texture units from it. In fact, in terms of stream processors, TMUs and rasterization devices, they are identical to the previous generation, and the main job of AMD engineers was to optimize the core and eliminate the weaknesses of Cypress discovered over the year. One of them was tessellation blocks: during this time, several games have appeared that actively use this technology, there is a tendency to an increase in their number, and therefore the frankly low performance of Cypress in this task has been corrected: according to AMD, Barts in optimal modes is twice as good as Radeon HD 5870 for this indicator.

At the same time, the characteristics of the Radeon HD 6800 are clearly inferior to their predecessors: if the Radeon HD 5870 had 1600 processors and 80 texture units, then the HD 6870 has only 1120 and 56. The ratio between the younger models is similar: Radeon HD 6850 boasts only 960 SP and 48 TMUs versus 1440 and 72 in the Radeon HD 5850. The number of ROPs and the capacity of the built-in caches have remained unchanged, and the weakening of the shader domain of the GPU AMD is trying to neutralize by increasing the frequency and by internal optimizations. Also note that Barts uses a GDDR5 controller from the Redwood core (Radeon HD 5600), which takes up half the space used in Cypress. As a result, the number of transistors in the new GPU decreased from 2.15 billion to 1.7 billion, and the core area - from 334 mm2 to 255 mm2, which, in turn, was reflected in the TDP: for the Radeon HD 6870 it is 151 W, for the HD 6850 - 127 W, and in idle both models consume and emit only 19 W. Barts is manufactured using TSMC's 40nm process technology - its long debugging forced AMD to abandon the design of these GPUs for 32nm, while the next generation will be produced according to 28nm standards.

Unlike the computing part of the GPU, other units have undergone much more serious modernization. First, the video processing module has been significantly improved: the third generation of the UVD engine - UVD3 - has the ability to hardware decode several new codecs. First, added the ability to play MVC (Multiview Video Coding) - another addition to the H.264 / AVC standard, which describes the transmission in one data stream of two frames with different angles required for stereoscopy. As a result, Northern Islands fully supports hardware playback of Blu-ray 3D and other 3D video compressed with this codec. The second major innovation was the introduction of full support for decoding and processing MPEG-4 ASP, the most common representatives of which are the DivX and XviD codecs. On the one hand, they are already perfectly reproduced in software and do not require serious resources, on the other hand, now even HTPCs with a weak CPU and a new AMD video card (here we are talking more about future budget models of the new generation) will be able to play video in this format, and laptop PCs will also save battery power. Finally, UVD3 now fully supports MPEG-2, including entropy coding algorithms, which was not available in previous Radeon models.

There have been major changes in the supported graphical interfaces as well. First, the Radeon HD 6800 is equipped with HDMI 1.4a, bringing support for FullHD (1080p24) stereoscopic video. Secondly, support for DisplayPort 1.2 has been introduced, which now looks more advantageous than HDMI and DVI, for several reasons. Firstly, the new version of the interface has a doubled bandwidth (21.6 Gb / s), which makes it possible to transmit a signal over one channel to 2 monitors with a resolution of 2560 × 1600 or 4 with 1920 × 1200 and a frequency of 60 Hz. If this is not important for video, then for games in 3D this means that now you can connect an output device not only via Dual-Link DVI (HDMI is not suitable for this use, since it has a limitation on the frequency of the transmitted signal: either 2x1080p at 24 Hz , or 2x720p @ 120Hz). Also, it becomes possible to transmit a high-resolution signal with a greater color depth (up to 30 bits). The capabilities for HD video playback have also been improved: now, via DisplayPort, you can transmit full-fledged uncompressed eight-channel audio in LPCM format simultaneously with FullHD video, as well as a lossless stream in DTS Master Audio and Dolby TrueHD codecs (they previously lacked bandwidth) ...

Also, the fact that DisplayPort does not require a clock generator for synchronization and is a packet, made it possible, with an increase in the bus width, to realize the ability to connect multiple monitors to a single video card output either by daisy chain or through a hub. The packets simply indicate which of the connected devices a particular frame is intended for, and the hub "parses" the stream into its components, sending each monitor the appropriate signal. For AMD, this means a significant simplification of the Eyefinity infrastructure - no longer requiring the rare and expensive Eyefinity6 model to connect six displays to a single graphics card. The Radeon HD 6800 implements the ability to build such a configuration using only two mini-DisplayPort (three monitors to each through hubs). Unfortunately, there are no DP 1.2-capable displays or hubs on the market yet - in this case, AMD is ahead of the curve - but they should be presented soon (most likely at CES 2011 in January).

Also with the release of Barts video cards, AMD finally presented its own version of stereoscopic image display technology, called HD3D. However, in this case, the solution is purely software and has nothing to do with the architecture of new products. Moreover, it is realized through additional drivers from third-party developers - TriDef and iZ3D. Among other improvements, we note the corrected anisotropic filtering, which now does not depend on either the angle or the nature of the textures, and the appearance of the Morphological Anti-Aliasing mode, which allows you to smooth the edges of contrasting objects with less performance loss than with regular supersampling (in fact, it's just DirectCompute- a filter, which is applied to a scene that has already passed rasterization, finds contrast zones characteristic of the edges of objects and smoothes them).

As you can see, the first models of the new generation of AMD video cards are a kind of compromise: on the one hand, they are much simpler and clearly weaker than the previous ones, on the other hand, they are much cheaper. Whether the balancing is successful will be shown by the test results.

The reference AMD Radeon HD 6870 and HD 6850 look very similar to their predecessors: they use similar cooling systems with a turbine fan and a plastic casing that completely covers the boards. The visual differences lie in the video interface connectors: now two DVI and one HDMI complement not the only full-size DisplayPort, but two mini-DisplayPort. In addition, the Radeon HD 6850 has only one six-pin power connector, not two; moreover, AMD got rid of the decorative protrusions on the casing, which sometimes prevented connecting connectors to them. Note that the older modification will initially be available exclusively in the form of copies of the reference version (apparently, TSMC's yield of suitable crystals is not too high yet), but the younger novelty already exists in many versions with coolers and printed circuit boards modified by several vendors.

First of all, we checked how much the tessellation performance of the new AMD video cards has been increased. The test carried out in Unigine Heaven shows that as the complexity increases, the Radeon HD 6800 does indeed turn out to be noticeably more productive than its predecessors: already at the Normal level, the Radeon HD 6870 outperforms the formally faster HD 5870, and the HD 6850 comes out on to Extreme. NVIDIA is not talking in this case: firstly, the Unigine engine is clearly faster on video cards from this vendor (as can be seen from the mode with disabled tessellation), and secondly, Fermi with its Polymorph Engine blocks is in any case much faster than Barts and Cypress in this task.

As for the more real tests, here we see that AMD has really turned out to be very successful products. In various applications, the ratio of forces changes somewhat, but in general, the following tendency can be noted: the maximum gap between the Radeon HD 6870 and HD 5870 does not exceed 15% and is sometimes reduced to a few percent. On average, among our measurements, the top-end video card of the previous generation is only 7% ahead of the new product, with a formal superiority in the number of execution units and a real difference in price by almost one and a half times. We see a similar success when compared with NVIDIA's direct competitor - the GeForce GTX 460 with 1 GB of memory: the Radeon HD 6870 runs 4-27% faster, depending on the test. Note also that the more expensive GeForce GTX 470 is not much ahead of the new AMD product, and the prices for it in Ukrainian retail will not soon approach those recommended by NVIDIA after the $ 260 drop. True, there are overclocked versions of the GeForce GTX 460 on the market, and the user himself can easily add performance to this video card, but at best we can only talk about parity.

As for the AMD Radeon HD 6850, this model also looks more than convincing in its niche: with a price difference of about $ 50, it is only 15% inferior to the Radeon HD 5850, and by about the same amount ahead of the cheaper NVIDIA GeForce GTX 460 by only $ 10 with 768 MB of memory. Here, obviously, there will be more intense competition than between senior modifications: the balance of forces and prices are quite close, besides, support for PhysX, CUDA and 3D Vision plays on NVIDIA's side.

conclusions

The new, or, more precisely, the updated graphics architecture of AMD is a pronounced manifestation of not revolutionary, but evolutionary development. The company did not continue the trend of annual leaps in productivity, instead, it brings existing opportunities closer to the average consumer. The main trump card of the Radeon HD 6800 is not the speed, but its ratio to the price, and by this factor the new items are head and shoulders above their predecessors and are in sharp competition with NVIDIA GeForce GTX 460, which we called the "kings" of the middle segment back in the summer.

With the release of these accelerators, AMD satisfies the needs of the bulk of demanding video card consumers, and the company will delight the most prestigious (and smallest) category of them very soon with the release of the Radeon HD 6900.

Positioning among AMD graphics cards

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

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

The next slide demonstrates the positioning of the new graphics 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, the AMD Radeon HD 6990 accelerators will replace the ATI Radeon HD 5970, leading the highest tier of AMD products. Below will be the video cards AMD Radeon HD 6950 and Radeon HD 6970, while the performance of the Radeon HD 6900 should significantly exceed the current line of ATI Radeon HD 5850 and Radeon HD 5870 accelerators. New items in the person of AMD Radeon HD 6800 actually replace the ATI Radeon HD 5800 series. ATI Radeon HD 5770 will remain on the market for now, 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 commensurate with the 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, the NVIDIA GeForce GTX 470 and GeForce GTX 480 accelerators will be opposed by solutions based on the yet to be presented AMD "Cayman" GPUs. 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.

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

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

	HD 5770	GTX 460 768MB	HD 6850	HD 5830	GTX 460 1GB	HD 6870	HD 5850	GTX 470
GPU	Juniper XT	GF104	Barts pro	Cypress	GF104	Barts xt	Cypress Pro	GF100
Technical process	40 nm	40 nm	40 nm	40 nm	40 nm	40 nm	40 nm	40 nm
Number of transistors, mln.	1040	1950	1700	2154	1950	1700	2154	3200
Shader units	800	336	960	1120	336	1120	1440	448
TMU	40	56	48	56	56	56	72	56
ROPs	16	24	32	16	32	32	32	40
GPU frequency	850 MHz	675 MHz	775 MHz	800 MHz	675 MHz	900 MHz	725 MHz	607 MHz
Memory size / type	1024 MB GDDR5	768MB GDDR5	1024 MB GDDR5	1024 MB GDDR5	1024 MB GDDR5	1024 MB GDDR5	1024 MB GDDR5	1280MB GDDR5
The width of the memory interface	128 bit	192 bit	256 bit	256 bit	256 bit	256 bit	256 bit	320 bit
Memory frequency	1200 MHz	900 MHz	1000 MHz	1000 MHz	900 MHz	1050 MHz	1000 MHz	837 MHz
Recommended price	$140	$160	$180	$190	$200	$240	$260	$260

Introduction

The eternal confrontation between the "reds" and "greens" has been going on for many years, and the situation on the fronts of this war continues to remain tense, despite temporary, albeit rather long, periods of calm - after all, they are always replaced by new bloody battles. AMD's all-encompassing reign in the DirectX 11-capable discrete graphics sector is still in our memory, but more recently - by industry standards - Nvidia was finally able to complete the transfer of most of its product lines to the new Fermi architecture. But not even a month has passed, and we will again have to witness another duel between the giants of the gaming 3D graphics market - the Radeon HD 6800 is entering the arena.

The onslaught of the graphics division of Advanced Micro Devices, formerly ATI Technologies, is sometimes overwhelming. In less than six months since the announcement of the first DirectX 11 graphics core, the ATI team has brought 11 graphics cards to market, from the humble Radeon HD 5450 to the mighty Radeon HD 5970, still the world's fastest single graphics card. In fact, AMD did not have much need to update the Radeon HD lines, but the company has learned well the lesson about the dangers of resting on its laurels; in addition, the retaliation from Nvidia in the form of the GeForce GTX 460 turned out to be significant enough to think about a symmetrical response as soon as possible. Not least of all, this was influenced by the situation with the performance of modern GPUs when performing tessellation: it is in this area that Nvidia has already demonstrated a significant advantage.

As we said in one of our previous reviews, the launch of the Nvidia GeForce GTX 460 family has become a serious threat to AMD, which could shake its dominance in the sector of so-called "people's gaming cards" - solutions that are simultaneously available to a significant percentage of buyers and at the same time sufficient productive to run modern games with a comfortable level of performance. Until recently, the Radeon HD 5830 and Radeon HD 5850 reigned in this segment, but the former is too truncated in configuration, uses an expensive printed circuit board, and the Cypress core itself was originally created for use in a higher price segment. As for the Radeon HD 5850, it is good for everything, except for the price. Thus, AMD urgently needed an adequate response to the threat from the Nvidia GF104, and this is partly why the company decided to start announcing the new generation of Radeon HD, also known as Northern Islands, with mass solutions, which is not quite usual, since the flagships are usually announced first.

At the moment, the AMD Radeon HD generation change strategy is presented as follows:

It is quite obvious that the number 8 in the name of the new line will no longer mean belonging to the most powerful single-processor solutions - now this privilege is marked with the number 9. The core of AMD's new "main battle tank" is the core codenamed Barts:

In the process of developing a new mainstream chip, AMD's main efforts were not focused on achieving maximum performance at any cost, which Nvidia often sins: Barts was created with an eye to the optimal combination of price, speed and functionality in its price range. And although at the same time not a new 40-nm technical process was used, the Barts developers were able to increase the packing density of the elements, which, coupled with a reduction in the number of transistors, made it possible to make the novelty compact, profitable in production, but possessing very serious technical characteristics and boasting a number of interesting innovations. ...

Radeon HD 6800: a place in the family

The developments of ATI Technologies, which later became part of Advanced Micro Devices, were often truly revolutionary and were often ahead of their time, which, however, did not go to their advantage. Can we say the same about the new Radeon HD family, which has changed the most significant number in the name from 5 to 6? Let's try to understand this issue.

At first glance, the new solutions from AMD based on the Barts core even represent a slight step back in comparison with the Radeon HD 5800 family: the number of ALUs and texture processors has decreased, as well as both fillrates. The new Barts is simpler and smaller than Cypress both in terms of the geometric area of ​​the crystal and the number of transistors included in its composition. If we adhere to such a superficial approach to the end, we can say that the Radeon HD 6800 has only a higher core clock speed of the older model, reaching 900 MHz versus 850 MHz for the Radeon HD 5870. In other quantitative indicators, Barts is inferior to Cypress.

However, this approach is fundamentally wrong. Firstly, due to its superficiality as such - and we know that the architecture of modern graphics processors is very complex and the performance can depend on the organization of shader processors much more strongly than on the direct number of ALUs. Secondly, we should not forget that the previous generation chip, Cypress, was developed as the most productive solution with an acceptable cost price, while Barts does not lead the Radeon HD 6000 family, but is positioned in the price sector, the lower limit of which is around 150 dollars, and the top does not exceed $ 250; in other words, Barts-based cards will have to compete mainly with Nvidia's solutions based on the GF104 - both in their current incarnation and, possibly, in future versions with unlocked 384 shader processors.

That is, if you look at Barts from the right angle, it does not at all look like a step backward from the Radeon HD 5800, but rather, it is a giant leap forward compared to the Radeon HD 5700 and the most dangerous rival of the GeForce GTX 460. AMD Barts core outperforms Nvidia GF104 in all parameters, while being simpler and more economical, at least at first glance. And, of course, in no case should we forget about the innovations, which are a lot in the new AMD graphics processor; in any case, enough to justify the number 6 in the name of the new Radeon HD family. In general, even if you do not go into the details of the architecture of the Radeon HD 6800, but restrict ourselves to the basic technical characteristics, the new solutions from AMD look perfectly balanced. According to official comments from AMD, they aim to repeat the success of the Radeon HD 4850, which once set a new standard for performance in the class of not-too-expensive but powerful DirectX 10 gaming cards. At first glance, the Radeon HD 6850 and 6870 have every chance of repeating this impressive a feat in the DirectX 11 sector, thus becoming the new "people's cards", thanks to the developer's recommended prices - $ 179 and $ 239, respectively.

Since the architecture of the Radeon HD 6800 contains a number of innovations and improvements, we should tell you more about it.

Radeon HD 6800: Computing Processor Architecture

Despite the fact that a number of rumors circulated on the Web about a serious change in the architecture of computing VLIW-processors in the new Northern Islands family, in particular, that the developers abandoned the scheme "4 simple and 1 complex ALU per stream processor" (AMD prefers to call a similar stream core device) in favor of a simpler arrangement "4 identical ALUs per processor" that saves a fair amount of transistors, in fact, these assumptions were not confirmed. Barts is still based on the TeraScale 2 architecture, embodied in the Radeon HD 5000 family. The superscalar design of stream processors still provides for five ALUs per processor, and four of these ALUs were designed to execute simple instructions such as FP MAD, and the fifth with a more complex design can execute complex instructions - SIN, COS, LOG, EXP, and so on. In addition to the ALU, each computational processor also contains a branch control unit and an array of general-purpose registers.

The approach is interesting, but, to some extent, perhaps controversial, since in order to achieve maximum performance, it is necessary to load all five ALUs that make up such a processor, and this, in turn, requires meticulous optimization of the shader code and perfect operation of the thread manager. However, a tremendous amount of work to improve the latter has already been done in the design and implementation of the cores of the Radeon HD 5000 family in silicon, and as we already know from the results of numerous performance studies of this family, it was not done in vain.

Curiously, a second thread manager appears in Barts' flowchart. Given that only one Ultra-Threaded Dispatch Processor (UTDP) is depicted in the official Cypress diagram, one would assume that increasing the number of UTDP to two, one for each SIMD core array, was undertaken to further reduce downtime. processing power and optimization of the load of stream processors, which, coupled with the increased clock speed, should have provided Barts with the opportunity to fully compete with Cypress.

However, we managed to clarify this issue. The above block diagram of the RV870 was simplified, while in fact, Cypress also has two UTDP units, each of which is served by its own rasterizer. There is also a switch connecting them for optimal load distribution; this whole system, without any visible changes, migrated to Barts silicon. As for the rest, the layout of the new kernel has hardly changed. The basic unit at Barts is still the SIMD core, which includes 16 computational processors (80 ALUs in total). Each such core is serviced by its own logic, has its own local data share (its size, apparently, remained the same - 32 KB), a first-level cache of 8 KB, and is interfaced with four texture processors. The developers did not touch on the rather complex cache system, however, the number of SIMD cores in Barts itself was reduced, so its volume changed accordingly. At the moment, it is not known how many SIMD cores are physically included in the new processor, we only know that 14 SIMD cores are active in the Radeon HD 6870, and 12 in the Radeon HD 6850.

In pursuit of simplification, Barts' computational part has lost support for double-precision computing, which also suggests that the Radeon HD 6800 is more a development of the Radeon HD 5700 rather than a direct replacement for the Radeon HD 5800. This feature is likely to remain the prerogative of the more powerful Radeon HD 6900, which will be powered by an aggressively codenamed Cayman chip. Thus, the Radeon HD 6800 looks rather dubious as a GPGPU platform, at least for serious calculations. However, since programs for home consumers do not use the FP64 format, but rely on FP32, the lack of support for double precision computing will not affect the target audience of new products in any way.

Radeon HD 6800: second generation DirectX 11 tessellator

Tessellation has become a standard feature since the introduction of DirectX 11, but while the Radeon HD 5000 architecture met all the requirements of the new API, it was tessellation that was its weak point from the very beginning. We can say that this feature was implemented in the Radeon HD 5000 "for show". While Nvidia did not have solutions with DirectX 11 support in its arsenal, this did not pose a significant problem, especially since there were practically no games with tessellation support on the market, however, with the advent of the Fermi architecture, the situation changed, since solutions based on it had a significantly higher geometry processing speed, which was clearly seen in the Stone Giant and Unigine Heaven Benchmark benchmarks, as well as in the Metro 2033 game.

And if earlier tessellation was an interesting, but non-standard and practically unused opportunity by game developers, with the release of DirectX 11 it became the de facto industry standard, and in order not to lose to Nvidia in this area, AMD had to work on improving the tessellation unit in the new generation of Radeon HD ...

AMD already has 8 generations of tessellation technology, however, it would be more correct to say that the Barts core contains a DX11-compatible second-generation tessellation unit, since all generations "up to DirectX 11" can be ignored - they never found wide support by software developers.

Before we get into Barts' tessellation improvements, let's take a look at the entire DirectX 11 tessellation pipeline.

In short: the hull shader calculates the tessellation parameters for each face of the patch (ranges from 2 to 64), determining how many faces to split each; the tessellator calculates the coordinates of each new vertex; domain shader sends all information (texture coordinates, UVW coordinates, etc.) about all vertices down the pipeline. Optionally, the hull shader can convert the breakpoints for the triangular patch to the breakpoints for the square patch, so data can be transferred directly from the HS to the DS.

As you can see, the tessellation process is quite complex in itself, which, as a result, means that the ability of the tessellator itself to split primitives (patches) into several parts is not only one of the factors limiting performance.

The new second (or seventh, according to AMD) generation tessellation unit contains a number of improvements, but not for the entire tessellation pipeline. The developers have optimized flow control for domain shaders and changed the sizes of queues and buffers so that the peak performance of the new tessellator reaches its maximum precisely at relatively low tessellation levels. In other words, AMD is not in vain so actively warning about the dangers of excessive tessellation with a polygon size of less than 16 pixels - it looks like Barts tessellator reaches its peak performance at this (or larger) triangle size.

This kind of comment may be an attempt to devalue the lag of Norther Islands graphics processors with extremely aggressive tessellation from Fermi-based chips, which include many PolyMorph geometry engines. On the other hand, excessive tessellation in games can be detrimental, since the generation of each new triangle entails an increase in the calculation of color values, the number of texture fetches, etc. Modern graphics processors work with tiles of 2 * 2 pixels, that is, each polygon should preferably be 4, 8, 16, 32, 64 (and so on) pixels in size. As soon as the polygon becomes less than four pixels, there is a tremendous slowdown, since the GPU is actually forced to work with a large number of tiles. Thus, with a polygon size of one pixel, the performance drop in modern graphics processors can be catastrophic, and the gain in detail is almost imperceptible in real gaming conditions.

If you believe the official statements, the improvements made to the architecture of the Barts tessellator required a minimal increase in the number of transistors, but at the same time made it possible to achieve a twofold increase in the performance of this unit on some synthetic tasks. This statement, like any other, needs to be tested by practice. If the performance when performing tessellation really increased so significantly, moreover, not in synthetic, but in real tasks, then the Nvidia GeForce GTX 460 has only PhysX support and very specific software that uses the Nvidia CUDA platform instead of OpenCL or DirectCompute.

As for the "eighth generation" tessellators, it is the third in the correct classification of DirectX 11 - it will be implemented only in the Cayman (Radeon HD 6900), and here AMD promises a threefold increase in performance compared to Cypress. It is quite possible that in future chips AMD engineers will concentrate on increasing the performance of the tessellator itself, possibly on optimizing the operation of hull shaders. In future architectures - Sourthern Islands, Hecatonchires, etc. changes should be expected at the level of organization of the tessellation pipeline itself; for example, in the direction of what Nvidia Fermi offers, where each large array of stream processors has its own tessellator, which optimizes data streams.

Morphological AA - DirectCompute improves graphics quality

Among other innovations, it should be noted support for a new type of full-screen anti-aliasing - the so-called morphological anti-aliasing (Morphological Anti-Aliasing, MAA or MLAA).

The official presentation from AMD did not disclose the details of the new algorithm or any technical details of its implementation in the ATI Radeon GPU. However, information about it can be found in the corresponding publication (http://visual-computing.intel-research.net/publications/papers/2009/mlaa/mlaa.pdf) by Intel, which created it for smoothing traced images rays. We do not know how exactly such an algorithm is implemented in the Radeon HD 6800, however, the general principles of its operation are the same for the CPU and GPU.

According to the data in the publication, the MLAA algorithm finds certain structures in the rendered frame and mixes colors along the edges of these structures using certain rules depending on the angle of inclination, color and other features of the structures.
It would be logical to assume that these rules can be set from the driver or even directly by the program. As a consequence, they can continually improve over time.

The MLAA algorithm is somewhat similar to edge-detect CFAA, introduced back in the days of the Radeon HD 2900 XT, however, the significant difference is that MLAA detects not edges that are very different in color and located at certain angles, but captures all structures with different colors. located nearby, and determines the features of these structures. The most important difference is the fact that edge-detect CFAA uses pixel shaders, which essentially means loading the entire rendering pipeline, while MLAA uses compute shaders, which do not need to execute texture instructions and which use fewer data transactions.

MSAA 8x



MLAA 8x



MLAA 8x + SSTAA

The good news is that using MLAA 4x and MLAA 8x does not blur textures. The anti-aliasing quality provided by MLAA 8x is comparable to that of MSAA 8x on many surfaces with less performance degradation. Without a doubt, MLAA works on every facet.

Unfortunately, the new algorithm has a huge flaw: it does not work with translucent textures. For example, in the case of Fallout: New Vegas, you can see that the small details of the fence and tree branches are not smoothed, and some of the color information that can be seen when using MSAA is lost. This can be both a fundamental problem of the algorithm as a whole, and its specific implementation. Even the demos Intel created to demonstrate this technology used conventional hardware anti-aliasing for alpha textures, which are usually used to simulate vegetation and other objects rich in fine details. Therefore, to achieve the maximum quality of anti-aliasing when using MLAA, it is also required to activate anti-aliasing of transparent textures (TAA). As you can see in the corresponding screenshot, the quality of morphological anti-aliasing with TAA enabled is almost perfect. The MLAA 8x + supersampling TAA bundle is almost superior in quality to the MSAA 8x.

I must also say that MLAA support is not an exclusive feature available only to owners of Radeon HD 6800 - due to the use of DirectCompute 11 and local data share, the algorithm works on any other AMD graphics processor that meets the DirectX 11 specifications. In theory, there are no prohibitions and to execute it on the Nvidia Fermi platform.

Radeon HD 6800: New Anisotropic Filtering Algorithm

The improved anisotropic filtering algorithm deserves a mention:

Since anisotropic filtering no longer has a serious impact on the performance of modern GPUs, this allows the use of algorithms in which the quality of filtering does not depend on the angle of inclination of the plane. Both AMD and Nvidia have already switched to using high-quality anisotropic filtering, and in the case of the Radeon HD 6800, we are only talking about further improving the existing algorithm in order to "soften" the transitions between MIP levels, so that they are less noticeable on textures with a large number of small details.

Radeon HD 6800 series AFRadeon HD 5800 series AF

Unlike the situation with MLAA, the advantages of the new anisotropic filtering algorithm are clearly visible. Of course, in real games they will not be so clearly noticeable, but still, any more or less attentive player will see the difference, since there are a lot of similar scenes in modern games.

Thus, all of the above does not give a reason to talk about a "new AMD revolution" - the Radeon HD 6800 is not a cardinally new development, and, moreover, a "subversion of foundations", but is a systematic evolutionary development of the successful architecture of the Radeon HD 5800.

Radeon HD 6800: DP 1.2, HDMI 1.4a, Stereo-3D and Eyefinity for the Masses!

Until now, the Radeon HD 5000 display controller has been the most advanced on the market, providing unprecedented connectivity with three monitors per card, and up to six monitors in the dedicated Eyefinity6 Edition models. Considering that a similar unit, which is part of Nvidia's graphics cores, still allows the simultaneous connection of no more than two display devices, there was no particularly urgent need to modify the Eyefinity unit. Nevertheless, the display controller Radeon HD 6800 has received new functionality that makes it completely unattainable for a competitor. First of all, this is support for the DisplayPort 1.2 standard, which allows multi-stream data transmission.

In other words, any member of the Radeon HD 6800 family now supports the connection of six monitors at the same time, and some of them can be connected via the DisplayPort interface either in daisy chain mode or using a special switch.

There are no special restrictions on the configuration of the connected displays: it is permissible to use monitors with different interfaces and resolutions. In addition, DisplayPort 1.2 implements 120Hz refresh rate support for 3D stereo monitors. It is theoretically possible to connect 3D panels via HDMI, since the Barts video controller implements version 1.4a of this interface - however, in practice, at the moment, there are no monitors or TVs capable of operating in 120-Hz mode via HDMI.

Additionally, the display controller Radeon HD 6800 received a hardware color correction unit, which serves to correctly display colors when displaying images on monitors with extended color gamut. In fact, all of the above, coupled with the improved UVD3 video processor, makes the Radeon HD 6800 the most advanced multimedia solution on the market. At least in theory.

Radeon 6800: Universal Video Decoder 3.0

The new, third version of the Unified Video Decoder video processor is interesting primarily because the already implemented support for decoding H.264 and VC-1 formats has been added full hardware support for DivX / XviD decoding, as well as support for entropy decoding for the MPEG-2 format. In addition, the chip is able to decode HD video in the Adobe Flash 10.1 format. Support for hardware decoding of Blu-ray 3D is also announced, but this is not as clear-cut as it looks in the presentation.

Formally, the ability to simultaneously decode two video streams in 1080p format, required by the Blu-ray 3D standard, is also implemented in the Radeon HD 5800/5700/5600/5500 video processors. However, in practice, everything turns out to be somewhat more complicated. The fact is that although the MPEG4-MVC codec is based on MPEG4-AVC (H.264), when decoding, it is necessary to take into account the dependence of the two visible frames on each other. In other words, despite the fact that cards of previous generations can simultaneously decode two streams of 40 Mbps each, they do not know how to hardware synchronize them to obtain a three-dimensional effect. Obviously, software synchronization is quite possible, however, as AMD humbly hints, the UVDs of previous generations were “not qualified” to decode and play Blu-ray 3D, which in practice could mean the company’s reluctance to update software and / or BIOS for HD 5000 series products. ...

AMD also claims that the Radeon HD 6800 is able to score 198 points in the HQV 2.0 test with a maximum result of 210 points, but this loud statement needs to be verified, as well as whether the new product outperforms solutions based on the Radeon HD 5000 architecture in this test.

Like its predecessors, the Radeon HD 6800 fully supports secure audio streams and can deliver 7.1-channel audio (192 kHz 24-bit) at up to 6.144 Mbps in AC3, DTS, Dolby True HD, DTS HD / DTS HD Master Audio formats. LPCM (Linear Pulse Code Modulation) and others via HDMI for further decoding by an external receiver.

As mentioned above, all the innovations do not make the new AMD graphics core revolutionary - they only complement and expand the capabilities originally laid down in the design of the Radeon HD 5000 architecture.

On this note, you can complete the theoretical part of today's review and move on to the practical one - acquainting readers with the material incarnations of the new generation of Radeon HD. By tradition, let's start with the older model.

Radeon HD 6870 PCB Design and Cooling Design

Even outwardly, the new generation of Radeon HD differs significantly from the old one - a strict, chopped design with sharp corners replaced the smooth contours and rounded corners. It cannot be said that the new design of the cooling system casing influenced anything, however, it is impossible to confuse the Radeon HD 6870 with the Radeon HD 5870 or HD 5850 under any circumstances, moreover, the new product is one and a half to two centimeters longer than its predecessor:

Radeon HD 6870Radeon HD 5850

Unlike the Radeon HD 5870, the Radeon HD 6870 lacks a metal heatsink on the back of the PCB. This part of the novelty looks quite ordinary, and no interesting design features deserving a separate mention were found here, except for one CrossFire connector versus two in the Radeon HD 5800 family. Of course, the most interesting is hidden inside. After dismantling the cooling system, the following picture appeared to our eyes:

The first thing that catches your eye is the non-standard, to say the least, the layout of the power subsystem. The four-phase GPU power regulator is located not in the tail of the PCB, as usual, but in the front, just behind the DVI, HDMI and DisplayPort connectors. It is built using integrated assemblies that combine power MOSFETs and their drivers. Perhaps such a strange arrangement was chosen in order to increase the efficiency of cooling the power elements, but, one way or another, such a solution has never been encountered in our practice before.

The heart of the GPU power regulator is the CHL8214 controller from CHiL Semiconductor. These controllers are quite rare on board modern graphics cards - until now we have known the only case in the person of Nvidia GeForce GTX 480. According to the technical description, CHL8214 is the oldest model in the line.

Memory power management is handled by the humble uP6122 chip from uPI Semiconductor. It and its accompanying power elements are located in a more familiar place on the printed circuit board, in the same place as the connectors for connecting external power. Both connectors are six-pin, with a recommended maximum load of 75W, and given the simpler design of Barts compared to the RV870, they should be enough to power the Radeon HD 6870, despite the increased voltage of the graphics core to 1.175V. The developers were forced to increase it in order to ensure stable operation of the graphics processor at a frequency of 900 MHz. The design of the printed circuit board does not provide for the possibility of installing eight-pin power connectors with increased load capacity.

If the design of the Radeon HD 5870 used memory chips manufactured by Samsung Semiconductor, then the H5GQ1H24AFR chips manufactured by Hynix are installed on board the Radeon HD 6870. The chips have a capacity of 1 Gbit (32Mx32) and are designed for a supply voltage of 1.5 V, and the T2C suffix in the marking indicates the nominal frequency of 1250 (5000) MHz. There are eight of them installed on the board; thus, the total volume of the local video memory bank is the standard 1024 MB today. With a 256-bit access bus at a frequency of 1050 (4200) MHz, the memory subsystem of the Radeon HD 6870 has a peak bandwidth of 134.4 GB / s, which is practically the same as the GeForce GTX 470. The lack of memory bandwidth in the Radeon HD 6870 clearly does not suffer will.

The Barts crystal has an unusual rectangular shape and is significantly smaller than the RV870. The heat-spreader cover is not used in the GPU design, as in all ATI / AMD solutions; protective measures are limited by the presence of a metal frame on the crystal packaging. For the first time in the history of the Radeon family, there is no engraving with the ATI logo on the crystal surface - the AMD logo now flaunts in its place, since, as we already know, Advanced Micro Devices made a decision (in our opinion, very rash) to abandon the ATI brand. The tradition of marking incomprehensible to an ordinary user, however, is fully preserved - from it one can only glean the date of manufacture of a given batch of crystals. In our case, this is the 36th week of 2010, which fell on the beginning of September, that is, by that time, AMD already had at its disposal solid lots of Barts, capable of operating at 900 MHz.

The GPU-Z utility version 0.4.7 already knows how to work with Barts and correctly recognizes the configuration of the new graphics chip, with the exception of the revision number. The absence of a tick in the OpenCL checkbox is due to the fact that the tests used the regular version of AMD Catalyst drivers, and not the APP Edition that adds OpenCL support. The only noticeable drawback of GPU-Z is that the utility does not display the number of texture processors, but their number corresponds to the official specifications for the Radeon HD 6870 - 56 TMU. Another favorite utility of enthusiasts, MSI Afterburner, also quite correctly detects new Radeon HD solutions, but in version 2.0.0 it does not yet know how to control the voltage of the graphics core. The diagnostic panel clearly shows that in the power saving mode, the GPU frequency decreases from 900 to 100 MHz, and the memory frequency drops to 300 (1200) MHz. This should provide high efficiency in modes that lightly load the GPU.

As mentioned, the new Radeon HD family has unprecedented connectivity. Indeed, as many as five connectors have settled on the mounting plate: a pair of DVI-I and Mini DisplayPort ports and an HDMI connector. Judging by the marking, only the bottom DVI-I port provides the possibility of analog connection through the appropriate adapter. As for the DisplayPort ports, they support the DP ++ mode, that is, they can emulate the operation of the DVI interface when an inexpensive passive adapter is connected. The configuration of the monitors connected to the Radeon HD 6800 can be practically any, as it was described in the theoretical part of the review. As regards CrossFire support, the new cards have only one connector, and, apparently, combining more than two Radeon HD 6800s is not supported. Most likely, this feature is reserved for the more powerful Radeon HD 6900.

The design of the cooling system has not undergone fundamental changes, and there are no revolutionary innovations in it. An aluminum plate, equipped with thermal pads in the right places, is responsible for cooling the memory chips and power elements of the power system, and an aluminum radiator on a copper base removes heat from the graphics core.

The radiator has a rather modest heat transfer area, but it is equipped with three heat pipes at once, two of which have a diameter of 8 millimeters. The heatsink is not mechanically connected to the aforementioned frame in any way and is attached to the board by means of four spring-loaded screws and a cruciform elastic plate, which ensures a reliable clamping of the base to the crystal. At the point of contact, a layer of dark gray thermal paste is applied. The photo clearly shows the profiling aerodynamic ribs of the casing, directing part of the air flow towards the side wall of the casing, since the space on the mounting plate for the ventilation slots is limited due to the large number of connectors. This is not to say that the described design makes an impressive impression, but taking into account the fact that Barts is simpler than Cypress, it should have a lower heat dissipation level, which means that such a cooling system should be enough for it, despite the increased core voltage. The only question is the comfort of the acoustic characteristics.

Radeon HD 6850 PCB Design and Cooling Design

The younger model of the new family is somewhat shorter than the older one, however, the power connector is located not on the top side of the board, but on the end side, so with the connected cable the dimensions of the Radeon HD 6870 and Radeon HD 6850 can be considered the same. The cooling system shroud is made in the same chopped style.

Both the front view and the rear view do not reveal anything interesting to the researcher, at least until the cooling system is dismantled. Like the older model of the new family, the younger one has only one CrossFire connector.

Unlike the Radeon HD 6870, the Radeon HD 6850 uses a conventional PCB layout with the power subsystem in the tail. Despite the reduced clock speed and GPU voltage, the power supply is also built in a four-phase scheme.

The same controller is responsible for its operation as in the older model - CHL8214 manufactured by CHiL Semiconductor.

The element base of the memory power regulator, in which the uP6122 microcircuit is used, is completely the same. This part of the power subsystem is located at the front of the printed circuit board. The power connector for the Radeon HD 6850 is only one and also six-pin, which means that the load on the power section of the PCI Express slot promises to be much higher than in the case of the Radeon HD 6870, which is partially compensated by the lower core voltage in 3D mode - 1.05 V versus 1.175 V. The design of the board does not provide for the possibility of installing an eight-pin connector.

As memory, the same microcircuits are used as in the design of the Radeon HD 6870 - Hynix H5GQ1H24AFR-T2C, capable of operating at a frequency of 1250 (5000) MHz. For the Radeon HD 6850, the use of such chips is firing a cannon at sparrows, since the standard memory frequency for this model is 1000 (4000) MHz. With a 256-bit access bus, these parameters provide a bandwidth of 128 GB / s. The total size of the local memory bank is 1024 MB. In power saving mode, the memory frequency is automatically reduced to 300 (1200) MHz.

The GPU die marking looks a little different than in the case of the Radeon HD 6870. The last line is made in a different font, while the first line, indicating the manufacturing time, contains the letter U. Unfortunately, we can only guess what it means. We only know for sure that this instance of Barts was made a week later than the one described above, installed in our instance of Radeon HD 6870.

The core configuration is determined correctly, we only add that the Radeon HD 6850 has only 48 texture processors active out of the 56 physically available ones. Just like in the previous case, MSI Afterburner does not know how to control the voltage of the graphics core, but at least it shows that the energy saving technologies work correctly: the GPU frequency in idle time is reduced to 100 MHz, and the memory frequency - to 300 (900) MHz. As a reminder, the core of the Radeon HD 6850 does not need to operate at ultra-high frequencies, so its supply voltage is lowered to 1.05 V.

The configuration of connectors in the younger model of the Radeon HD 6800 family is the same as in the older one: the card carries on board a pair of DVI-I and DisplayPort ports with DP ++ support and multi-streaming, as well as an HDMI port that meets 1.4a specifications. Complementing this splendor is the only CrossFire connector, which allows you to combine a pair of Radeon HD 6850s into a single multi-GPU tandem; asymmetric configurations with the Radeon HD 6870 are most likely supported.

In general terms, the design of the Radeon HD 6850 cooling system resembles the design of the Radeon HD 6870 cooler described above, however, it is much simpler: the radiator has a significantly smaller heat transfer area and is equipped with a single flat U-shaped heat pipe at the base. The dimensions of the radiator do not inspire respect at all. As with the Radeon HD 6870, the shroud has aerodynamic fins that direct some of the airflow towards the side cover of the system case.

An additional element of the cooling system is a figured plate with low ribbing, which removes heat from memory microcircuits and power stabilizer assemblies, for which there are heat-conducting pads in the right places on it. This plate is attached to the board separately from the heatsink and plastic shroud. This cooling system does not seem capable of any serious feats, especially since its design uses a less powerful and more compact fan, however, and the graphics core of the Radeon HD 6850 works in less stressful conditions than its twin installed in the Radeon HD 6870. We will try to find out how efficient the cooling systems of the new Radeon HD family are in the next chapter of our review.

Power consumption, thermal conditions, noise and overclocking

The electrical characteristics of any new graphics solution are of serious interest, and we always pay close attention to this aspect. The new Radeon HD models have not passed the traditional testing - they were subjected to the standard testing procedure on the measuring platform with the following configuration:

Intel Core 2 Quad Q6600 Processor (3 GHz, 1333 MHz FSB x 9, LGA775)
DFI LANParty UT ICFX3200-T2R / G (ATI CrossFire Xpress 3200) Motherboard
PC2-1066 memory (2x2 GB, 1066 MHz)
Power supply unit Enermax Liberty ELT620AWT (power 620 W)
Microsoft Windows 7 Ultimate 64-bit
CyberLink PowerDVD 9 Ultra / "Serenity" BD (1080p VC-1, 20 Mbps)
Crysis warhead
OCCT Perestroika 3.1.0

This stand is equipped with a special measuring module described in the review “ Power consumption of computers: how many watts do you need?". Its use allows you to get the most complete data on the electrical characteristics of modern graphics cards in various modes. As usual, the following tests were used to load the video adapter in various modes:

CyberLink PowerDVD 9: FullScreen, Hardware Acceleration Enabled
Crysis Warhead: 1600x1200, FSAA 4x, DirectX 10 / Enthusiast, frost card
OCCT Perestroika GPU: 1600x1200, FullScreen, Shader Complexity 8

For each mode, with the exception of the OCCT limit load simulation, measurements were taken for 60 seconds; In order to avoid card damage due to overloading of power circuits, for the OCCT: GPU test, the test time was limited to 10 seconds. Using this technique, we were able to obtain the following results:

As expected, the Radeon HD 6870 turned out to be significantly more economical than the Radeon HD 5870, but the increased GPU voltage was not in vain for it - in 3D mode, the power consumption level turned out to be almost the same as in the Radeon HD 5850. But in modes, where the load on the core is not too strong, the efficiency of the new item is much higher. The load on the +3.3 V power line turned out to be unexpectedly high, which is practically not used in modern graphics cards for quite a long time. As for the rest, the behavior of the Radeon HD 6870 in terms of power consumption is quite predictable; in particular, from the very beginning we assumed approximately equal load on the power connectors. And so it turned out; the slight overweight attributable to the connector, indicated in the table as "12V 6/8-pin", can be disregarded.

With the Radeon HD 6850, the picture is more interesting: numerous repeated measurements in 2D mode invariably yielded results in the region of 30-33 W, while the core frequency, according to MSI Afterburner, did indeed drop to the prescribed 100 MHz. Apparently, incorrect PowerPlay operation was observed in the pre-sale sample of the card that fell into our hands; for example, in idle mode, the system could not reduce the GPU supply voltage, which led to an increased level of power consumption in the absence of a real load. The same applies to workloads like decoding high-definition video - the result also turned out to be higher than that of the Radeon HD 6870. But in 3D mode, where the core voltage is maximum, correct results were obtained. Here, the Radeon HD 6850 consumes significantly less than its counterpart, which is quite natural given the lower frequency, lower supply voltage, and fewer active GPU units. The nature of consumption along individual lines of the Radeon HD 6850 is similar, however, due to the presence of only one power connector, this single connector is loaded much more and in the synthetic OCCT test the power consumption through this channel reaches 80 watts.

So, from the point of view of efficiency indicators, the new Radeon HD family turned out to be very successful, except for an unpleasant failure in the PowerPlay logic of the Radeon HD 6850 in some modes, but this behavior is unlikely to be observed in serial cards supplied to retail chains. But even with this amendment in 3D mode, the younger model consumes a little more than the much more modest Radeon HD 5770 in terms of performance. As for the older model, it is at least as efficient as the Radeon HD 5850, being, according to AMD promises, faster than the latter in modern games. Not a bad claim to be the leader in its class, especially since the Nvidia GeForce GTX 460 1GB is a significantly less economical solution.

The new Radeon HD models demonstrate a very intense thermal mode of operation, which is not least due to the not very productive reference cooling systems. The merit is dubious, but, in fairness, it should be noted that most reference coolers of powerful graphics cards differ in this behavior, while non-standard systems often demonstrate much more impressive performance. Thus, the coolness of the Radeon HD 6870 and Radeon HD 6850 does not differ, but this is true only for the reference versions of these cards. They are likely to be followed by solutions equipped with better cooling systems. In addition, values ​​in the region of 75-80 degrees Celsius have long been the norm for modern GPUs, and in no way should you be afraid of them.

With regard to the noise level, the situation is ambiguous: if, in the absence of a serious load, the new Radeon HD 6800 models behave very quietly, practically merging with the background noise of the operating system (38 dBA for a test laboratory), then when running resource-intensive applications that actively use the graphics processor, their fans quickly increase the speed and the cards become clearly audible. The younger model of the family, according to the sound level meter, is somewhat quieter than the older one, but there is no noticeable difference by ear, at least according to our feelings. This is not to say that the noise level is too high - after all, any productive gaming cards make a lot of noise, but it should be understood that purchasing a Radeon HD 6870 or Radeon HD 6850, you will not get a silent solution in all modes, at least if refers to models equipped with a reference cooling system.

Investigating the HD Video Playback Capabilities of the Radeon HD 6800

The already traditional improvement of the UVD engine with each new generation makes it clear that the developers are positioning the AMD Radeon HD 6800, including for HD video lovers. Let's see how good the Barts GPU is for multimedia tasks in theory and practice.

So, UVD 3.0 allows hardware decoding of streams in DivX / XviD, MPEG2-HD, MPEG4-AVC, MPEG4-MVC, WMV-HD, VC-1, Adobe Flash 10.1 and a number of others. In addition, it supports the transfer of many audio formats over HDMI, as well as hardware post-processing of SD and HD video. In other words, the UVD 3.0 video engine is not much different from its predecessor and is its logical evolutionary development.

At first glance, it seems rather strange to introduce support for DivX / XviD hardware decoding and add entropy decoding support for MPEG2 in 2010. However, it should be understood that UVD 3.0 was primarily developed not only for graphics cards with a maximum consumption of more than 100 W, but for further integration into various mobile graphics or central processors. When decoding video, the consumption of UVD 3.0 should be less than that of the higher performance CPU. It is only surprising that when playing HD video, the Radeon HD 6850 consumes almost 40 W: not a very serious load for a desktop system, but significant for a mobile one.

Obviously, for a desktop PC owner, power consumption is hardly as important as such. A low volume of the cooling system and a generally comfortable acoustic level are required (alas, the reference Radeon HD 6850 is not a really quiet graphics card), but an equally important component is the quality of video playback, both HD in native resolution and SD when interpolated to 1080p resolution. ...

In this part of our article, we'll take a look at how well UVD 3.0 and Radeon HD 6850 can decode Blu-ray discs as well as play high definition video and interpolate standard video to FullHD.

Test platform configuration and test methodology

The study of the quality and performance of Nvidia GeForce GTX 460 and other graphics processors when playing and decoding video streams was carried out on a test system with the following configuration:

Intel Core 2 Duo processor E8500 (3.16 GHz, 6 MB cache, 1333 MHz bus)
Gigabyte EG45M-DS2H Motherboard (Intel G45)
OCZ Technology PC2-8500 memory (2x1 GB, 1066 MHz, 5-5-5-15, 2T)
Western Digital Hard Drive (640GB, SATA-150, 16MB buffer)
Antec Fusion 430W Chassis
Monitor Samsung 244T (24 ”, maximum resolution [email protected] Hz)
Optical drive LG GGC-H20L (Blu-ray, HD DVD, DVD)
ATI Catalyst 10.6 / 10.9 / 10.10 for ATI Radeon
Nvidia ForceWare 197.45 / 258.96 / 260.63 / 260.99
CyberLink PowerDVD 10
Microsoft Windows Performance Monitor
Microsoft Windows 7 64-bit

The following graphics cards took part in the study:

AMD Radeon HD 6850
ATI Radeon HD 5750
ATI Radeon HD 5670
ATI Radeon HD 5570
ATI Radeon HD 4770
Nvidia GeForce GTS 450
Nvidia GeForce GTX 460
Nvidia GeForce 9800 GT / GTS 240
Nvidia GeForce GT 240

The following tools were used to assess the quality of video playback in standard (SD) and high (HD) definitions:

IDT / Silicon Optix HQV 2.0 DVD
IDT / Silicon Optix HQV2.0 Blu-ray

Driver settings remained unchanged. However, in accordance with the requirements of the HQV test suite, the noise reduction levels and detail enhancements in the drivers were increased to medium (50-60%), which did not affect the results of multi-cadence tests.

Given the interest of owners of expensive sound systems in the results of playback of uncompressed audio streams, we included DTS-HD Master Audio and Dolby Digital TrueHD (where available) to increase the load on the central processor in all reproduced samples.

Taking into account the fact that tests are carried out on the Windows 7 operating system without disabling background services, jumps in the maximum level of CPU utilization should not be taken critically. The most important are the average parameters of the level of occupied processor time. As a result, it makes sense to remember that a difference of 1-2% does not mean an unambiguous advantage or disadvantage of this or that accelerator in comparison with a competitor.

The following films were used to assess the processor load when playing FullHD video (1920x1080), as well as FullHD video with activated picture-in-picture (BonusView in the Blu-ray disc Association classification):

Alien Vs. Predator ": MPEG2 HD Part 18
Constantine: VC1 Picture-in-Picture Part 25
Dark Knight: VC1 Part 1 (excluding credits)
Death Race: MPEG4-AVC / H.264 Picture-in-Picture Part 14
"The Day After Tomorrow": MPEG4-AVC / H264 Part 14

Video playback quality

HQV 2.0 test packages make it possible to subjectively assess the quality of execution of a number of video processing operations by a graphics processor. As already mentioned, the test is very detailed and focused on comparing Blu-ray / DVD-players (built on the basis of specialized video processors), as a result of which modern GPUs are far from always capable of showing really high results.

HQV 2.0 DVD

The specificity of the current situation in the video market is such that few people watch ordinary DVD-movies on TVs with a "native" DVD resolution, and more and more on screens with FullHD resolution (1920x1080). Thus, the main task of the video processor is not so much the correct display of content, but the ability to perform high-quality interpolation, correct motion, reduce noise, increase the clarity of details, and so on. The video excerpts presented in the HQV 2.0 DVD are aimed precisely at understanding how well modern chips can perform the above operations separately.

AMD didn’t say anything about improving image quality when it announced UVD 3.0. As you can see, it is not in vain: the interpolation quality of the Radeon HD 6850 is fully consistent with its predecessors.

HQV 2.0 Blu-ray

Much like the HQV 2.0 DVD, the HQV 2.0 Blu-ray test suite allows you to subjectively examine similar video processor capabilities at high resolutions.

As in the previous case, we do not see a single difference from the test results of our predecessors, which is not bad on the whole. The results of the Radeon HD 5000/6800 are traditionally higher than the competing solutions Nvidia GeForce and most of its shortcomings (test results with 0 points) relate to low quality content. Users who watch HD movies from Blu-ray discs, rather than trying to stretch a pseudo-HD image from iTunes or similar services to full screen, are unlikely to be dissatisfied with the image quality on the Radeon HD 6800.

With the release of the Radeon HD 6850 series and Catalyst 10.10 drivers, AMD began to set the noise removal and edge enhancement settings to a rather aggressive default level. We are at a loss to say why this was done, but it is clear that this will maximize the results of the corresponding test videos in HQV 2.0. Unfortunately, AMD's custom noise reduction technology is far from perfect, even at 50% it does not eliminate noise artifacts as much as blur the picture, as a result of which many 720p videos look literally like VHS tapes.

Considering the fact that real films contain many scenes shot in different places with different lighting and sometimes with different cameras, the value of video processors is precisely in the ability to customize themselves for a specific scene on the fly. Therefore, we would recommend that users check the default noise reduction and sharpness settings in the drivers.

Interestingly, the HQV 2.0 Blu-ray benchmark did not work on a Radeon HD 6850 graphics card without being updated to the latest version. At the same time, all films were played excellently. A new version of Cyberlink PowerDVD 10 with support for AMD Radeon HD 6800 and Blu-ray 3D is due out this month.

When considering the results of the HQV tests, it should be remembered that the scoring method is extremely subjective and therefore the small difference between the final scores of different cards can hardly be considered critical.

Blu-ray playback

Consider how well the Radeon HD 6800 is able to offload the system's CPU from decoding high-definition video.

There are no particular changes in the playback of the films "Dark Knight" and "Constantine" in the novelty: it shows very good, but not outstanding results.

The average CPU load when playing our MPEG4-AVC movies for the Radeon HD 6850 is at a very decent level - about 7%. Moreover, the maximum indicators are also slightly reduced, which reduces the possibility of jerking during playback.

Based on the data obtained, GPU decoding of MPEG2 HD entropy noticeably reduces the average and maximum CPU load times. As you can see, the HD 6850 is a clear leader among the Radeon series in this respect.

Multimedia capabilities: what is the bottom line

Like most of its predecessors, the AMD Radeon HD 6850 chip is an exceptional home theater graphics card.

Supports hardware decoding of video streams in DivX / XviD, MPEG2-HD, MPEG4-AVC, MPEG4-MVC, WMV-HD, VC-1, Adobe Flash 10.1 and a number of others, having the ability to transfer all common types of audio formats via HDMI 1.4a, and with quality SD and HD video post-processing hardware, the AMD Radeon HD 6850 is the most advanced card on the market when it comes to multimedia capabilities. Unfortunately, the Radeon HD 6850 consumes a lot of power and is rather bulky, so you shouldn't hope for such passively cooled graphics cards. The HD 6870 is so long that it won't fit into any reasonably sized HTPC case.

The Radeon HD 6850 has better Blu-ray playback and DVD interpolation than competing solutions in its class, but still not perfect according to HQV 2.0. Apparently, the developers will have to modify the Avivo engine in the chip or the drivers to show significantly better results in the HQV 2.0 tests.

It should be noted separately that the technology for 3D stereo image output - AMD HD3D - supports the output of Blu-ray 3D movies to a very wide range of TVs and projectors without the need to purchase additional software (except for a Cyberlink PowerDVD Deluxe player with support for Blu-ray 3D). In the case of competing 3D Vision, you also need to purchase a dedicated driver from Nvidia.

Test platform configuration and performance testing methodology

Testing of the new Radeon HD 6800 models in conditions as close as possible to real life was carried out on a universal test platform with the following configuration:

Intel Core i7-975 Extreme Edition Processor (3.33 GHz, 6.4 GT / s QPI)
Cooler Scythe SCKTN-3000 "Katana 3"
Gigabyte GA-EX58-Extreme motherboard (Intel X58)
Memory Corsair XMS3-12800C9 (3x2 GB, 1333 MHz, 9-9-9-24, 2T)
Samsung Spinpoint F1 Hard Drive (1TB / 32MB SATA II)
Ultra X4 850W Modular Power Supply (850 Watts Nominal)
Dell 3007WFP Monitor (30 ”Max Resolution [email protected] Hz)
Microsoft Windows 7 Ultimate 64-bit

The following versions of ATI Catalyst and Nvidia GeForce drivers were used:

ATI Catalyst 10.10a (with hotfix) for ATI Radeon HD
Nvidia GeForce 260.89 WHQL for Nvidia GeForce

The drivers themselves were configured as follows:

ATI Catalyst:

Anti-Aliasing: Use application settings / Standard Filter
Morphological filtering: Off
Texture Filtering Quality: High Quality
Surface Format Optimization: Off
Wait for vertical refresh: Always Off
Anti-Aliasing Mode: Quality

Nvidia GeForce:

Texture filtering - Quality: High quality
Vertical sync: Force off
Antialiasing - Transparency: Multisampling
CUDA - GPUs: All
Set PhysX configuration: Auto-select
Ambient Occlusion: Off
Other settings: default

The test package includes the following games and applications:

3D first-person shooters:

Aliens vs. Predator (1.0.0.0, Benchmark)
Battlefield: Bad Company 2 (1.0.1.0, Fraps)
Call of Duty: Modern Warfare 2 (1.0.182, Fraps)
Crysis Warhead (1.1.1.711, Benchmark)
Far Cry 2 (1.03, Benchmark)
Metro 2033 (Ranger Pack, 1.02, Benchmark)
S.T.A.L.K.E.R .: Call of Pripyat (1.6.02, Fraps)

3D shooters with a third person view:

Just Cause 2 (1.0.0.1, Benchmark / Fraps)
Lost Planet 2 (1.1, Benchmark)

RPG:

Mass Effect 2 (1.01, Fraps)

Simulators:

Colin McRae: Dirt 2 (1.1, Benchmark)
Tom Clancy "s H.A.W.X. (1.03, Benchmark)
Tom Clancy "s H.A.W.X. 2 (1.01, Benchmark)

Strategy games:

BattleForge (1.2, Benchmark)
StarCraft II: Wings of Liberty (1.0.2, Fraps)

Semi-synthetic and synthetic tests:

Futuremark 3DMark Vantage (1.0.2.1)
Final Fantasy XIV Official Benchmark (1.0.0.0, Fraps)
Unigine Heaven Benchmark (2.0)

Each of the games in this suite of test software has been tuned to provide the highest possible level of detail. Applications that support tessellation have used this feature.

A fundamental refusal to manually modify any configuration files means that only the means available in the game itself to any uninitiated user were used for customization. The tests were carried out at resolutions 1600x900, 1920x1080 and 2560x1600. Except where otherwise noted, the standard 16x anisotropic filtering was complemented by 4x MSAA anti-aliasing. Anti-aliasing was activated either by means of the game itself, or, in their absence, it was forced using the appropriate settings of the ATI Catalyst and Nvidia GeForce drivers.

In addition to the Radeon HD 6870 and Radeon HD 6850, the following graphics cards took part in the tests:

ATI Radeon HD 5870
ATI Radeon HD 5850
Nvidia GeForce GTX 470
Nvidia GeForce GTX 460 1GB
Nvidia GeForce GTX 460 768MB

To obtain performance data, we used the testing tools built into the game with the obligatory use of the original test videos, and, if possible, recording the data on the minimum performance. In the absence of the aforementioned tools, the Fraps 3.2.3 utility was used in manual mode with a three-fold test pass, fixing the minimum values ​​and then averaging the final result.

Game Tests: Aliens vs. Predator

The improved tessellation block is performing well. Of course, the new Radeon HD 6800 will not be able to reach the GeForce GTX 470 with all the desire, but the older model quite successfully reaches the level of the GeForce GTX 460 1GB, and in resolutions from 1920x1080 it is ahead of it in the minimum performance; however, only indicators in 1600x900 can be called more or less comfortable. Thanks to architectural improvements, even the Radeon HD 6850 outperforms the Radeon HD 5870 in this game. But that's just the beginning.

Game tests: Battlefield: Bad Company 2

The results are in good agreement with AMD's claims. With a smaller number of functional units, the Radeon HD 6870 successfully competes with the Radeon HD 5850, however, this merit is almost entirely due to a serious difference in the frequency of the processors of these graphics cards. The youngest model of the new family, Radeon HD 6850 successfully surpasses its plan, outperforming the GeForce GTX 460 768MB and reaching the level of the GeForce GTX 460 1GB. Considering the lower price point, this makes the Radeon HD 6850 a very attractive solution. But so far this is only the second gaming test, but what will happen next?

Game tests: Call of Duty: Modern Warfare 2

In the third test, the Radeon HD 6870 was able to fulfill AMD's promises - to show the same as the Radeon HD 5850 - only at 1600x900, and starting from 1920x1080 it began to lag behind the Radeon HD 5850 more and more. Fortunately, the average and minimum indicators remained at a comfortable level even at 2560x1600. Given the different price ranges, hardly anyone would seriously want to change the ATI Radeon HD 5850 to the AMD Radeon HD 6850, given the fact that there are not so many games that use tessellation. Nevertheless, it is significant that the 6800 series is sometimes slower than the 5800.

Game tests: Crysis Warhead

This game, despite the weight of its engine, does not use tessellation, so there is nowhere for Barts to fully reveal their talents. As a result, the older model of the new family is content with the role of the heir to the Radeon HD 5850, and the younger one very successfully competes in high resolutions with the GeForce GTX 460 1GB. Not bad, but taking into account the exactingness of the game, it makes no sense from a practical point of view - close to acceptable indicators are demonstrated by cards of this class, except in 1600x900 resolution.

Game tests: Far Cry 2

Interestingly, despite the 900 MHz core frequency, the Radeon HD 6870 begins to lag behind the Radeon HD 5850 as the resolution grows, and at 2560x1600 this lag reaches 7%, which may indicate insufficient memory bandwidth; Fortunately, we are talking only about average performance, and the minimum does not change, and in general, the speed reserve for both cards is enough to provide acceptable conditions for the player. The fate of the Radeon HD 6850 in this case is competition with the cheaper GeForce GTX 460 768MB, and even then, at 1600x900 it does not do it very well. However, the 2560x1600 resolution is also available for the younger model of the new Radeon HD 6800 family.

Game tests: Metro 2033

This game is tested without anti-aliasing. Tessellation is enabled.

The use of a new test with tessellation enabled makes it possible to understand the exactingness of Metro 2033. Even at 1600x900 resolution, only the GeForce GTX 470 manages to show over 40 frames per second, at a minimum speed of no more than 12 frames per second, that is, one can only dream of completely comfortable conditions. As for the Radeon HD 6870, the superiority in the minimum performance over the Radeon HD 5850, which is about 1-3 frames per second, is completely insufficient to objectively judge the capabilities of the new tessellation unit or other optimizations in Barts.

Once again, we can state that the Radeon HD 6800 is slower than the Radeon HD 5800.

Game tests: S.T.A.L.K.E.R .: Call of Pripyat

In this test, DX10.1 and DX11 modes are used for cards with the corresponding capabilities. Tessellation is enabled.

In another post-apocalyptic shooter, the new products manage to show more or less the same performance as the Radeon HD 5000. Bearing in mind that in S.T.A.L.K.E.R .: Call of Pripyat tessellation is used very conditionally, it cannot be said that the new chips show their potential power here. Quite the opposite: a large number of execution units of the Radeon HD 5800 successfully compete with the high frequencies of the Radeon HD 6800.

The AMD Radeon HD 6870 manages to maintain performance on par with the GeForce GTX 460 1GB, which is officially priced at $ 40 less, which is not a convincing position. The junior representative of the new line looks good, showing the speed similar to the GeForce GTX 460 768MB.

Game tests: Just Cause 2

The integrated testing tools do not display minimum performance information, so we use Fraps to get it.

Tessellation in Just Cause 2 is not implemented, however, the option to simulate the behavior of water surfaces using GPU forces is used. The core of the Radeon HD 6870 runs at 900 MHz, which is reflected in the geometry processing speed accordingly. Even if the architectural improvements in Barts affected only the tessellation block, without affecting other blocks related to geometry processing, the difference in such frequency alone is enough to achieve in this game performance almost on the level of the Radeon HD 5870. Taking into account the price difference of the Radeon HD 6870 and the Radeon HD 5870 is a great result. The Radeon HD 6850 also feels good, but it does not set any records anymore, being content with parity with the GeForce GTX 460 768MB in the first two resolutions and providing an opportunity to play comfortably at 1600x900.

Game tests: Lost Planet 2

The advantages of Barts when performing tessellation are clearly visible: at 1600x900, the Radeon HD 6870 outperforms even the Radeon HD 5870 in minimum performance.On the other hand, 22 frames per second looks like a breakthrough only against the background of the Radeon HD 5850, while the GeForce GTX 460 768MB is easy provides the same, and its brother, equipped with 1 GB of video memory, generally maintains a minimum speed at a level close to 30 frames per second, which is beyond the power of either the younger or even the older model Radeon HD 6800.

Game tests: Mass Effect 2

In this test, full-screen anti-aliasing is forced using the technique outlined in the Contemporary Graphics Accelerators in Mass Effect 2 review.

Both Radeon HD 6800 models demonstrate impressive results, especially at 2560x1600 resolution, in which a fairly high minimum speed is demonstrated only by them and the more expensive (officially - $ 259) and hot GeForce GTX 470. The Radeon HD 5800 family cannot boast of such a thing, despite its superiority over the Radeon HD 6800 family in a number of technical characteristics. Its minimum performance can be called conditionally acceptable, but they do not reach 25 frames per second.

Game tests: Colin McRae: Dirt 2

For cards supporting DirectX 11, the corresponding mode is used. Tessellation is enabled.

Despite the new tessellation unit, the Radeon HD 6800 family does not perform as brilliantly in this test as in some others, simply because tessellation speed is not a bottleneck in this game. Here the older model naturally competes with the Radeon HD 5850, and not at all with the Radeon HD 5870. The younger representative, the Radeon HD 6850, unfortunately, is quite inferior to both versions of the Nvidia GeForce GTX 460, with the exception of the 2560x1600 resolution, where it manages to achieve parity with the GeForce GTX 460 768MB. However, the lag behind the GeForce GTX 460 1GB is minimal, and the overall performance level demonstrated by the Radeon HD 6850 is quite sufficient for the practical use of this resolution.

Game tests: Tom Clancy's H.A.W.X.

For testing, the tools built into the game are used, which do not provide for fixing the minimum indicators. Uses DirectX 10 / 10.1 modes.

In the first part of H.A.W.X. The new Radeon HD models again prove that they are not in vain classified as the next generation - in particular, the Radeon HD 6870 easily overtakes the GeForce GTX 460 1GB at 1920x1080 and even the GeForce GTX 470 at 2560x1600, and this test has always been considered "Nvidia's territory". The Radeon HD 6850 is not so successful, but even starting from the 1920 × 1080 mode, it is quite capable of competing with cards based on Nvidia GF104.

Game benchmarks: Tom Clancy's H.A.W.X. 2 Preview Benchmark

Before publishing the results of the preliminary H.A.W.X. 2 we have to make a reservation that this application was distributed by Nvidia up until October 22nd, 2010.

This test uses tessellation to draw the surface of the earth. Tessellation increases the number of primitives to 1.5 million per frame, excluding planes, trees and buildings, while the size of a typical primitive is 6 pixels, which is very suboptimal from a number of points of view.

Preliminary test H.A.W.X. 2 (not the game itself, which has not yet been released) returns the undisputed leadership to Nvidia solutions. Yes, the Radeon HD 6870 is ahead of the Radeon HD 5870, and quite significantly, but, despite the improved tessellation unit, it is far from even the GeForce GTX 460 768MB, not to mention the more powerful Fermi solutions. The only consolation is the good absolute performance of the new products, which allows you to play even at 2560x1600.

It should be noted that the preview benchmark H.A.W.X. 2 is heavily criticized by AMD, which claims that this "preview product" does not deliver performance comparable to other tessellation applications. In particular, according to some Internet resources, AMD claims the following:

“It has come to our attention that you may have received an early build of a benchmark based on the upcoming Ubisoft title H.A.W.X. 2. I "m sure you are fully aware that the timing of this benchmark is not coincidental and is an attempt by our competitor to negatively influence your reviews of the AMD Radeon HD 6800-series products. We suggest you do not use this benchmark at present as it has known issues with its implementation of DirectX 11 tessellation and does not serve as a useful indicator of performance for the HD 6800 series.A quick comparison of the performance data in HAWX 2, with tessellation on, and that of other games / benchmarks will demonstrate how unrepresentative HAWX 2 performance is of real world performance.

AMD has demonstrated to Ubisoft tessellation performance improvements that benefit all GPUs, but the developer has chosen not to implement them in the preview benchmark. For that reason, we are working on a driver-based solution in time for the final release of the game that improves performance without sacrificing image quality. In the meantime we recommend you hold off using the benchmark as it will not provide a useful measure of performance relative to other DirectX 11 games using tessellation. "

AMD's irritation is understandable enough that H.A.W.X. The 2 preview benchmark uses tessellation beyond measure, making it the main performance bottleneck. It is quite interesting to observe that H.A.W.X. 2 benchmark is faster than the real H.A.W.X. game, and you can draw certain conclusions based on this.

Game tests: BattleForge

For cards supporting DirectX 11, the corresponding mode is used.

Alas, the problem with the minimum performance of the Radeon HD has not disappeared, even in the new generation based on the Barts core. Although the average performance of the Radeon HD 6870 and Radeon HD 6850 is quite high, the minimum speed is below any criticism, while at 1600x900 even the GeForce GTX 460 768MB is able to maintain this parameter at a level of at least 30 frames per second.

Game Tests: StarCraft II: Wings of Liberty

The main achievement of the Radeon HD 6800 in this test is a rather serious breakthrough in the minimum performance, especially in comparison with the Radeon HD 5850. Moreover, in the resolution of 1920 × 1080 the older model of the new family managed to bypass even the GeForce GTX 470. But the resolution of 2560 × 1600, alas, is the same. remained closed due to insufficiently high minimum indicators, although the Radeon HD 6870 came close to the cherished 25 frames per second.

Semi-synthetic and synthetic benchmarks: Futuremark 3DMark Vantage

To minimize CPU impact, 3DMark Vantage uses an “Extreme” profile that uses 1920x1200 resolution, 4x FSAA and anisotropic filtering. For the sake of completeness, the results of individual tests are taken over the entire resolution range.

The Radeon HD 6870 managed to overcome the 8,000 point bar, at least in the overall standings. The final result turned out to be even higher than the result of the GeForce GTX 470. But the Radeon HD 6850 did not quite reach the level of the GeForce GTX 460 1GB, although it was ahead of its younger brother.

In the second test, the Radeon HD 6800 family shows itself much better than in the first, especially the older model. Since the performance of the geometry engine is important in this test, the result is quite natural. But, as we already know from the results of game tests, this is far from enough for a confident victory over the rivals from the “green” team.

Semi-synthetic and synthetic benchmarks: Final Fantasy XIV Official Benchmark

Since initially the FF XIV Official Benchmark produces a meaningless result in glasses, Fraps is used to obtain data on the performance of graphics cards. The test only supports resolutions of 1280x720 and 1920x1080.

Testing has shown nothing new: this test is still the domain of the Radeon HD, where it reigns supreme. Let's just note that the Radeon HD 6870 is not inferior to the Radeon HD 5870 in the resolution of 1920x1080, not being its direct rival.

Semi-synthetic and synthetic benchmarks: Unigine Heaven benchmark

The test uses tessellation in "normal" mode.

Despite the strengthened tessellation unit, the Radeon HD 6800 family did not show a dramatic improvement in the results in this test, except that the older model was able to bypass the minimum performance of the Radeon HD 5870 in the resolution of 1920 × 1080. limited by other factors? In any case, the promised breakthrough in this test did not work, but the results shown by the Radeon HD 6800 cannot be a failure either.

Radeon HD 6870: advantages and disadvantages

Advantages:

High level of performance in modern games
May outperform Radeon HD 5870 in some tests

Wide range of FSAA modes






HDMI 1.4a support
DisplayPort 1.2 support

Disadvantages:

Noticeable noise level

Radeon HD 6850: pros and cons

.
Advantages:

Decent performance in its class
Fast tessellation performance compared to Radeon HD 5800
Wide range of FSAA modes
Industry leading anisotropic filtering performance
Support for output to six monitors
Full hardware support for HD video decoding, including DivX and 3D
High-quality post-processing and HD video scaling
Integrated audio engine with support for HD audio formats
Supports HDMI audio output
HDMI 1.4a support
DisplayPort 1.2 support
Low energy consumption for its class
High efficiency in energy saving modes

Disadvantages:

In low resolutions it is inferior to GeForce GTX 460 768MB
Noticeable noise level
Not too efficient cooling system
Smaller choice of GPGPU-accelerated software compared to competing solutions

Conclusion

So, we have tested the new Radeon HD 6800 family in 19 different gaming and synthetic tests. What can you tell by looking at the results of these tests?
In general, the older model AMD Radeon HD 6870 performed very well: in most cases it is faster than the more expensive ATI Radeon HD 5850, while it has a number of improvements, including better performance of the tessellation unit, which was evident in several tests. Pivot charts illustrate this well.

It should be noted that in the resolution 1600x900 the struggle with the GeForce GTX 460 1GB lasted with varying success, but already in 1920x1200 the new AMD product began to lead quite confidently, and in 2560x1600 the average superiority of the Radeon HD 6870 over the competitor reached 16%. Moreover, in most tests, the Radeon HD 6870 not only showed performance at the level of the Radeon HD 5850, but also outperformed it, in some places quite significantly. In fact, this is a verdict on the latter, as, in fact, is planned by Advanced Micro Devices itself. However, given the price of the Radeon HD 6870, for those looking for an inexpensive but powerful graphics card for use in modern games, it makes sense to look at the GeForce GTX 460 1GB, especially the versions with factory overclocked to 750-800 MHz at the core frequency. This solution will prove to be no worse in practice than the Radeon HD 6870, and in addition, will provide the player with support for minor improvements like PhysX in a number of games. As for the owners of the Radeon HD 5870, they do not need to worry for now, at least until the announcement of the Radeon HD 6900.

Things are more complicated with the Radeon HD 6850. It is inferior to its older brother, on average, by about 15%, but in some cases the lag can reach 20-40%. Against the Radeon HD 5850, this new product also has no serious chances. Although where high speed is required when performing tessellation, the Radeon HD 6850 can quite seriously lead, but there are still few such games on the market. As for the rivalry with the GeForce GTX 460 768MB, there is reason for pessimism. Just look at the pivot charts.

At low resolutions, Nvidia's solution is definitely faster; Radeon HD 6850 wins only in a small number of tests and this gain is extremely insignificant. As the resolution increases, the situation evens out, however, in 1920 × 1080 the battle goes on with varying success, and here everything depends on the specific game, and the 2560 × 1600 mode was not originally intended for use in conjunction with cards of the Radeon HD 6850 or GeForce GTX 460 768MB class. Should you upgrade from Radeon HD 5830 to Radeon HD 6850? In our opinion, it is unambiguous that the new solution is much better balanced in terms of technical characteristics and performance. But if you choose between it and the GeForce GTX 460 768MB, you should be guided by a set of your favorite games.

In general, both models of the Radeon HD 6800 family should be recognized as successful, both in terms of price and in terms of technical characteristics and performance. The graphics development team at Advanced Micro Devices did a good job of fixing one of the bottlenecks of the Radeon HD 5800 architecture - low speed when performing tessellation and low overall speed of geometric information processing. In addition, a number of innovations related to the field of multimedia have made the new items truly unique. These innovations include support for DisplayPort 1.2, HDMI 1.4a, a new video processor that supports hardware decoding DivX, as well as the ability to connect up to six monitors or television panels, and in almost any configuration.

Considering the power consumption and dimensions of the Radeon HD 6850/6870, it is difficult to recommend such solutions for PC home theaters. However, when it comes to an HTPC aimed at gaming, the 6850 has the potential to be the top choice.

The Barts chip supports all possible high definition formats, including Blu-ray 3D, the highest, if not ideal, playback quality for Blu-ray content and DVD video interpolation, as measured by HQV 2.0 tests.

As a result, Nvidia, which at one time delayed the launch of its own architecture with DirectX 11 support, although it was able, in the end, to complete the transfer of its product lines to it, but did not get a long respite - by the time the company could Finally, to enjoy the fruits of Fermi, the former ATI Technologies has already prepared a new blow, and this blow turned out to be quite sensitive. Now all that remains is to wait for the announcement of the Radeon HD 6900 "Cayman" to see if it can regain AMD's leadership in creating the fastest single-processor graphics cards.

GeForce GTS 450 SLI: featherweight champion?

(or ATI Mobility Radeon HD 6830) is a high-performance laptop card that supports DirectX 11. Technically, this adapter has the same features as the Mobility Radeon HD 5830, but with a higher clock speed. Like all Mobility HD 5800 / 6800M series cards, the HD 6830M video adapter is based on the HD 5770 (RV840) desktop chip.

The HD 6830M memory interface consists of two 64-bit controllers that lead to a 128-bit memory bus. It provides access to 1024 MB DDR3 memory. Due to the narrow bus width and lack of GDDR5 support, memory performance can be considered the "weak link" of this card.

The card has 800 MADD-cores (so-called stream processors), which are collected in 160 5-dimensional groups. The cores support DirectX 11 hardware features (tessellation, OIT, post-processing, shadows, HDR texture compression) and have a theoretical compute power of 0.92 TFLOPS. In addition, 16 ROPs (rasterization and blending units), 40 TMUs (texture units) and 40 TAUs can be found on the chip. ATI Stream, OpenCL and DirectCompute 11 can help with general computing.

Gaming performance card 6830M lies between HD 5830 and 5850M / 6550M with DDR3 memory. This suggests that the GPU power for demanding games (Metro 2033, Call of Duty Black Ops or Mafia 2) will not be enough. You cannot play the above games at high settings. However, at a resolution of 1366x768 pixels, without AA (AntiAliasing - an anti-aliasing method that eliminates the "staircase effect") games should work fine.

Series Mobility Radeon HD 6800M Equipped with the old UVD2 video processor, which is used to decode HD video using a graphics card. It only "accepts" VC-1, H.264 and MPEG-2, while the new UVD3 in 6900M series adapters supports DivX. With Flash 10.1, the 6800M series can help accelerate Flash HD videos (YouTube).

Just like the HD 5830, the 6830M can carry eight-channel HD audio (Dolby True HD and DTS HD Master Audio) over HDMI 1.3a. The 6830M still supports the initial version of Eyefinity (not the improved Eyefinity +), which connects up to 6 monitors to the graphics chip. This is only possible if there is enough DisplayPort in the laptop.

The power consumption is about the same as the old one (about 24 W), so the adapter is installed in laptops with a diagonal of 15-17 inches.

Manufacturer:	AMD
Series:	Radeon HD 6830M [email protected]
Code:	Granville-LP
Streams:	800 - unified
Clock frequency:	575 * MHz
Shader frequency:	575 * MHz
Memory frequency:	900 * MHz
Memory bus width:	128 Bit
Memory type:	DDR3, GDDR3
Maximum memory:	1024 MB
Common memory:	No
DirectX:	DirectX 11, Shader 5.0
Energy consumption:	24 watts
Transistors:	1080 million
Technology:	40 nm
Laptop size:	big
Release date:	07.01.2010
Manufacturer link:	http://www.amd.com/us/products/notebook/graphics/amd-radeon-6000m/amd-radeon-6800m/Pages/amd-radeon-6800m.aspx#2

* The indicated clock speeds are subject to change by the manufacturer.