ATI Radeon HD 6800: Generation Next?

Advanced Micro Devices has just announced their new generation of ATI Radeon HD graphics architectures. Our today’s article will show whether they will turn out as successful as their predecessors.

by Alexey Stepin , Yaroslav Lyssenko, Anton Shilov
11/02/2010 | 08:10 AM

AMD and Nvidia have been competing for the graphics card market for many years, an occasional lull only meaning that the fighters are preparing for the next round. We can still recall how AMD was absolutely superior in the sector of DirectX 11 compatibles but then Nvidia finally transitioned nearly all of its product series to the modern Fermi architecture as well. And now we’re up to a new fight between the GPU developers as AMD rolls out its Radeon HD 6800 series.

 

AMD’s Graphics Products Group, formerly the independent company ATI Technologies, have amazing achievements under their belt. In less than a half-year since the announcement of the world’s first DirectX 11 compatible GPU, they released as many as 11 graphics card models from the entry-level Radeon HD 5450 to the top-performance Radeon HD 5970 which, by the way, still remains the fastest graphics card in the world. As a matter of fact, AMD didn’t particularly need to update this Radeon HD product line-up, yet resting on one’s laurels is never a good strategy. Besides, Nvidia had dealt quite a heavy blow with its GeForce GTX 460, so some kind of a response was called for. Another factor that might have worried AMD was that Nvidia’s solutions were better when doing tessellation.

As we wrote in an earlier review, the GeForce GTX 460 series is a serious threat to AMD in the sector of performance-mainstream graphics cards which combine two appealing qualities. They are affordable for the majority of users and they are fast enough to run modern games at a comfortable frame rate. This market sector used to be dominated by the Radeon HD 5830 and 5850 but the former has a cut-down configuration and uses an expensive PCB (and even the Cypress core itself had been designed for more expensive products). As for the Radeon HD 5850, it is good all around save for its price. Thus, AMD found itself lacking a good opponent to Nvidia’s GF104. This must be the reason why they are introducing their new Radeon HD generation, also known as Northern Islands, by announcing performance-mainstream solutions rather than top-end ones as is the customary practice.

Right now, AMD has the following strategy of changing its Radeon HD generations:

Clearly, the number 8 in the names of the new series products won’t mean top-end single-GPU solutions anymore. The number 9 is used for that purpose instead. The graphics core of these cards is codenamed Barts:

While developing the new GPU, AMD focused on striking an optimal balance of price, speed and functionality in the defined price range rather than on delivering sheer speed as Nvidia has been prone to do in its recent engineering efforts. Even though using the old 40nm tech process, the Barts features higher component density. Coupled with the reduced number of transistors, the new GPU turns to be compact and profitable to make, yet it also features impressive specs and a number of exciting innovations.

Radeon HD 6800 Series

Positioning

ATI Technologies, which is now part of AMD, used to come up with ingenious ideas and concepts that were often ahead of their time (and not always for their own good). Can we say the same about the new Radeon HD 6 series as compared to its processor? Let’s try to make everything out.


Click to enlarge

At first sight, the new Barts-based solutions seem to be a step backwards if you compare them to their Radeon HD 5800 counterparts. The reduced number of ALUs and texture-mapping units and the lower fillrate parameters may lead an inexperienced user to this conclusion. After all, ordinary users are often prone to be allured by big numbers and the Barts is indeed simpler and smaller than the Cypress, for example in its die size and transistor count. Superficially, the Radeon HD 6800 only seems to have but one advantage over the previous series. Its clock rate is 900 MHz (we mean the top model) as compared to the Radeon HD 5870’s 850 MHz. Otherwise, the Barts is inferior to the Cypress in every other quantitative parameter!

Well, of course this superficial approach to evaluating the new series is incorrect. First off, modern GPU architectures are so complex that the way their shader processors are organized may have a much higher effect on the GPU’s performance than the sheer number of ALUs it has. Second, we should keep it in mind that the previous-generation Cypress processor was devised as a top-performance chip with acceptable manufacturing cost whereas the Barts is not the top of the Radeon HD 6000 line-up. It is targeted at the price range of $150-250. In other words, Barts-based products are meant to compete with Nvidia’s GF104-based cards (both in the current as well as prospective versions with 384 shader processors).

So, it’s just a matter of perspective. Instead of viewing the Barts as a step backwards compared to the Radeon HD 5800, we must consider it as a great leap forward over the Radeon HD 5700 and as a dangerous rival to the GeForce GTX 460. The Barts core is superior to the GF104 in every parameter while being apparently simpler and more economical. Besides, we mustn’t forget about the host of innovations the new GPU brings about (there are enough of them to justify the number 6 in the model name, we must admit). Even if we don’t delve deep into the details of the Radeon HD 6800 architecture but limit ourselves to the basic specs only, we can see that the new solutions from AMD look very well balanced. According to official comments by AMD, they promise to be as successful as the Radeon HD 4850 which set a new standard of performance in the $199 segment of DirectX 10 compatibles in its time. The Radeon HD 6850 and 6870 have every chance to do the same among DirectX 11 compatible cards, especially as they have the right recommended pricing: $179 and $239, respectively.

As noted above, the Radeon HD 6800 architecture is full of various innovations and improvements. We’ll tell you more about them in the next section.

Execution Processors

Notwithstanding the earlier rumors about fundamental changes in the architecture of VLIW processors in the new Northern Islands family, particularly that the developer had given up the existing design with 4 simple and 1 complex ALU per a stream processor (AMD calls this device a stream core) in favor of a simpler design with 4 identical ALUs per processor, we don’t see such changes in the final product. The Barts is still based on the TeraScale 2 architecture that was introduced in the Radeon HD 5000 series. The same superscalar design still has five ALUs per a stream processor: four ALUs can execute simple instructions like FP MAD whereas the fifth one can execute complex instructions like SIN, COS, LOG, EXP, etc. Besides the ALUs, each stream processor contains a branch control unit and a set of general-purpose registers.

This approach is not unquestionable because achieving maximum performance calls for all the five ALUs in each processor to have some work to do, which requires shader code optimizations and an ideal operation of the ultra-threaded dispatch processor (UTDP). The latter had already been optimized extensively (and, judging by the resulting performance, successfully) for the Radeon HD 5000 series, though.

Interestingly, there is a second dispatch processor in the Barts flowchart. Considering that the official Cypress flowchart has only one UTDP, we might suspect that the two UTDPs, one for each SIMD array, are meant to further optimize the use of the available computing resources and, coupled with the increased clock rate, make the Barts competitive to the Cypress.

Well, we have managed to clarify the issue. As a matter of fact, the abovementioned RV870 flowchart was simplified. Instead, the Cypress has two UTDPs as well, each of which is serviced by a dedicated rasterization unit. The two are connected with a switch for optimal load balancing. This whole arrangement seems to have been carried over to the Barts silicon without any visible changes. The component layout of the new GPU hasn’t changed much in other ways, either.

The basic subunit of a Barts processor is a SIMD core which includes 16 stream processors (for a total of 80 ALUs). Each such core is serviced by dedicated logic and has a local data share (its size seems to have remained the same at 32 KB) and an 8KB L1 cache. It is connected to four texture-mapping units. The developer hasn’t revised the rather sophisticated system of caches but its total amount is different because the Barts has fewer SIMD cores than its predecessor. Right now, we do not know how many SIMD cores there are 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.

Following the simplification strategy, the Barts has been stripped of the support for double-precision calculations. By the way, this is yet another indication that the Radeon HD 6800 continues the Radeon HD 5700 series rather than replaces the Radeon HD 5800 one. Such calculations will probably be a prerogative of the more advanced Radeon HD 6900 which will feature a GPU known under the codename of Cayman. Thus, the Radeon HD 6800 doesn’t look appealing as a GPGPU solution for serious computing. On the other hand, applications for home users prefer FP32 to the FP64 format and their performance won’t be affected by the lack of double-precision support.

Second Generation DirectX 11 Tessellator

DirectX 11 made tessellation a standard feature but, even though the Radeon HD 5000 series complied with all the technical requirements of the new API, its tessellation performance was poor. We might even say that the Radeon HD 5000 cards offered but formal support for that feature. This was not a problem as long as Nvidia didn’t have DirectX 11 solutions at all, especially as there were almost no games utilizing tessellation. The release of the Fermi architecture changed the situation because Fermi-based cards were much faster than their Radeon HD opponents at processing geometry as we could see in the Stone Giant and Unigine Heaven benchmarks as well as in Metro 2033.

Tessellation had used to be an interesting but nonstandard and rarely employed opportunity until DirectX 11. But after it had become an industry standard, AMD had to work hard to improve its tessellation unit in the new Radeon HD generation and match Nvidia’s solutions in that respect.

AMD’s tessellation technology is eight generations old already, but we can disregard the six generations before DirectX 11 as they didn’t ever get much recognition from game developers. Thus, the Barts features a second-generation DirectX 11-compatible tessellation unit from AMD.

Before discussing what improvements have been implemented in the Barts in terms of tessellation, let’s take a look at the whole DirectX 11 tessellation pipeline.

Briefly, the hull shader calculates tessellation parameters for each patch edge and defines the number of fragments to split each edge into. The tessellator calculates the coordinates of each new vertex. The domain shader sends data (texture coordinates, UVW coordinates, etc.) about all vertexes along the pipeline. The hull shader can optionally convert the control points of a triangular patch into control points of a square patch. Therefore data can be sent directly from the hull shader to the domain shader.

 As you can understand from this brief description, tessellation is quite a complex process. It means that the tessellator’s ability to split primitives (patches) into multiple parts is one, but not the only, performance-limiting factor.

The new second-generation tessellation unit features a number of improvements, but not for the whole tessellation pipeline. The developers have improved thread management for domain shaders and re-sized some queues and buffers to achieve significantly higher peak throughput, particularly at lower tessellation levels. AMD warns that excessive tessellation with a polygon size below 16 pixels is harmful for performance, so we can infer that the Barts’ tessellator reaches its peak performance at that (or larger) triangle size.

This warning by AMD may be meant to explain why the Northern Islands GPUs may be inferior at very aggressive tessellation levels to the Fermi GPUs which incorporate numerous PolyMorph geometry engines. On the other hand, excessive tessellation can indeed be harmful because each new triangle involves more color calculations, texture sampling and other operations. Modern GPUs work with 2*2-pixel tiles, so it is desirable to have polygons the size of 4, 8, 16, 32, 64 pixels and so on. As soon as polygons become smaller than 4 pixels, the GPU has to process more tiles and slows down catastrophically. In other words, a modern GPU can suffer a terrible performance hit when processing 1-pixel polygons whereas the increased level of detail will hardly be noticeable during actual gameplay.

According to the official commentaries, the improvements in the Barts’ tessellator did not require much more transistors but helped make that unit twice as fast in some synthetic tasks. Like any other claim, this one needs checking out in practical applications, but if the new chip’s tessellation performance is indeed so high, the Nvidia GeForce GTX 460 will only have two technical advantages over its Radeon counterpart: PhysX and CUDA.

We can expect the future architectures, Southern Islands, Hecatonchires, etc., to bring some innovations into the very design of the tessellation pipeline, like what Nvidia offers in its Fermi architecture where each large array of stream processors has a dedicated tessellator for optimal data threading.

Morphological AA - DirectCompute for Higher Graphics Quality

Another notable improvement in the new GPU is its support for a new type of full-screen antialiasing called Morphological Antialiasing (MAA or MLAA).

AMD’s official presentation doesn’t reveal the details of the new algorithm or its implementation in the new Radeon processor but we can find them in an Intel publication where this antialiasing method was employed for images rendered using ray tracing. We do not know how exactly this algorithm is implemented in the GPU but its basic principle should be the same for both CPUs and GPUs.

So, the MLAA algorithm finds certain structures in a rendered frame and blends colors at the edges of such structures according to a set of rules. The blending depends on the angle of view, color and other features of the structures. We assume that such rules can be set by the driver or even by the application and thus can be improved over time.

The MLAA algorithm resembles the edge-detect CFAA that was introduced as far back as the Radeon HD 2900 XT but there is one important difference. Instead of detecting edges which differ greatly in color and are positioned at certain angles, MLAA finds all adjacent differently-colored structures and identifies their special features. And the most important difference is that the edge-detect CFAA method uses pixel shaders and thus the entire rendering pipeline whereas MLAA relies on compute shaders which do not call for texture instructions and involve fewer data transactions.

UPDATE: 4 November, 2010
 

MSAA 8x

MLAA 8x

MLAA 8x + SSTAA

The good thing about MLAA 4x and MLAA 4x is that they, as expected, do not blur textures. The quality of MLAA 8x is akin to MSAA 8x on many surfaces and it comes with lower performance hit. Indeed, the morphological antialiasing works on all angles.

Unfortunately, there is a huge drawback of MLAA: it does not work on alpha-textures. For example, in case of Fallout: New Vegas we see that the fence and the trees are still aliased and some of the color information that could be seen in case of multisampling (twigs, wires, etc) is missing. This seems to be a problem of the algorithm or its particular implementation itself. In fact, even Intel's demos use traditional hardware antialiasing for transparent textures usually used for vegetations and small items. Therefore, in order to make MLAA work perfectly, one would have to enable hardware transparency antialiasing (TAA). In fact, as you see on one of the screenshots, the quality of morphological antialiasing with TAA is simply perfect, in fact, MLAA 8x + supersampling TAA provides almost better quality than MSAA 8x!

 

By the way, MLAA is not an exclusive feature of the Radeon HD 6800 series. Based on DirectCompute 11 and local data shares, this algorithm can run on any other AMD GPU that complies with the DirectX 11 specs. There is no theoretical obstacle to using it on the Fermi architecture, either.

 

New Anisotropic Filtering Algorithm

The improved anisotropic filtering algorithm must be noted, too.

Anisotropic filtering doesn’t have a serious effect on the performance of modern GPUs, so algorithms can be used whose filtering quality doesn't depend on the angle of inclination of the plane. AMD and Nvidia have both transitioned to high-quality anisotropic filtering already and the Radeon HD 6800 just improves the algorithm further to smooth out transitions between mipmaps so that they would be less conspicuous in textures with lots of small details.

 


Radeon HD 6800 series AF


Radeon HD 5800 series AF

As opposed to MLAA, the benefits of the new AF algorithm are easy to see. They won't be so clear in the process of actual gaming, yet a sharp-eyed gamer will surely spot the difference, especially as modern games offer a lot of scenes suitable for that.

Thus, the Radeon HD 6800 doesn’t look like a completely new product. It is rather a steady and evolutionary development of the successful Radeon HD 5800 architecture.

DisplayPort 1.2, HDMI 1.4a, Stereo-3D and Eyefinity

Up till now the display controller of the Radeon HD 5000 series has been the most advanced one on the market, providing flexible connectivity options. Particularly, it allowed to connect up to three monitors to one graphics card (and special Eyefinity6 Edition products even supported up to six monitors simultaneously). Considering that Nvidia’s display controller supports no more than two displays at the same time, there has been no real need for improving the Eyefinity unit further. However, AMD has endowed the display controller of the Radeon HD 6800 series with new functionality to make it even more superior to its opponent. First of all, the new controller supports DisplayPort 1.2 for multiple-stream data transfers.

In other words any Radeon HD 6800 series card now supports up to six monitors simultaneously. Some of the monitors can be connected via DisplayPort in daisy-chain mode or using a special switch.

There are no special limitations as to the configuration of displays. You can use devices with different interfaces and resolutions. Besides, DisplayPort 1.2 supports a refresh rate of 120 Hz for 3D monitors. 3D panels can also be connected via HDMI because the Barts’ video controller complies with the HDMI 1.4a standard.

Additionally, the display controller of the Radeon HD 6800 series features a hardware color correction unit for displaying colors accurately on extended-gamut monitors. Summing everything up and adding the improved UVD3 video processor, the Radeon HD 6800 seems to be the most advanced multimedia solution available on the market right now, at least theoretically.

Universal Video Decoder 3.0

The third-generation video processor from AMD is interesting for its added support for hardware acceleration of DivX/XviD decoding and for entropy decoding of MPEG-2, which is added to the previously implemented decoding of H.264 and VC-1. The chip can also decode HD video in Adobe Flash 10.1 format. AMD claims the new video processor supports hardware decoding of Blu-ray 3D but that’s not as clear as it looks in the presentation.

The video processors of the Radeon HD 5800/5700/5600/5500 cards can formally decode two 1080p video streams simultaneously as required by the Blu-ray 3D standard, but things are somewhat more complicated in practice. Although the MPEG4-MVC codec is based on MPEG4-AVC (H.264), its decoding needs to take into account the relation between the two streams. In other words, although the previous-generation cards could decode two 40Mbps streams simultaneously, they could not synchronize them on the hardware to achieve the stereoscopic effect. Software-based synchronization is quite possible, though, but the previous-generation UVD was not qualified to decode and play Blu-ray 3D, which may mean that AMD just doesn't want to revise the software and/or BIOS for its HD 5000 series.

AMD also claims that the Radeon HD 6800 can score 198 points in the HQV 2.0 test, the maximum possible score being 210 points. The claim that the new solution is superior to the Radeon HD 5000 series in this respect needs checking out, of course.

Like its predecessors, the Radeon HD 6800 series supports Protected Audio Path and can bitstream 7.1 audio (192 kHz/24 bits per sample) with a bitrate up to 6.144 Mbps in AC3, DTS, Dolby True HD, DTS HD/DTS HD Master Audio, LPCM (Linear Pulse Code Modulation) and other advanced formats via HDMI 1.4a for further decoding on an external receiver.

Again, these innovations do not make the new graphics architecture from AMD something completely different from its predecessor. They only expand or add to the functionality of the Radeon HD 5000 series.

Now it’s time for us to finish the theoretical part of this review and move on to practice. We'll take a look at the new hardware starting from the senior model.

Radeon HD 6870: PCB Design and Cooling System

The new Radeon HD generation is much different from the older one. Instead of curvy lines and rounded-off edges, we now see a stern design with sharp corners. We wouldn’t say the new design of the cooler casing can affect anything in terms of performance, but it really makes it impossible to confuse a Radeon HD 6870 with a Radeon HD 5870 or HD 5850, especially as the new card is a couple of centimeters longer than its predecessor.

 


Radeon HD 6870


Radeon HD 5850

As opposed to the Radeon HD 5870, the Radeon HD 6870 lacks a metallic heat-spreading back-plate on the reverse side of the PCB. This part of the card is overall quite unexciting and lacks any interesting design features save for one CrossFire slots instead of two on the Radeon HD 5800 series. The most important things are hidden inside, of course. When we took the cooler off, we saw the following:

The first thing to catch our eye is the nonstandard layout of the power circuit. The four-phase GPU voltage regulator is located not in its usual place at the back of the PCB, but at the front, right next to the DVI, HDMI and DisplayPort connectors. It is based on integrated packs that combine MOSFETs and drivers. This unusual layout may help improve the cooling of the power components, but we must confess it’s the first time we’ve ever seen such a solution.

The GPU voltage regulator is based on a CHL8214 controller from CHiL Semiconductor. Such controllers are rather rare on modern graphics cards. We’ve only seen one on the Nvidia GeForce GTX 480. According to its product brief, the CHL8214 is a senior model in its series.

A small uP6122 chip from uPI Semiconductor is responsible for powering the graphics memory. It is located together with its accompanying chips in its traditional place near the external power connectors. Both connectors are of the 6-pin variety and have a load capacity of 75 watts. Considering the simplified design of the Barts processor compared to the RV870, they should be enough to power up this Radeon HD 6870 despite its higher GPU voltage (1.175 volts). The developer had to increase the latter parameter to make the GPU stable at a clock rate of 900 MHz. The PCB design does not provide for 8-pin power connectors with higher load capacity.

The Radeon HD 5870 used to come with memory chips from Samsung Semiconductor whereas the Radeon HD 6870 is equipped with H5GQ1H24AFR chips from Hynix. They have a capacity of 1 Gb (32 Mb x 32) and a rated voltage of 1.5 volts. The T2C suffix in the marking indicates their rated frequency of 1250 (5000) MHz. There are eight such chips on the PCB for a total of 1024 megabytes. The 256-bit memory bus and memory frequency of 1050 (4200) MHz provide a peak memory bandwidth of 134.4 GBps which is almost the same as that of the GeForce GTX 470. The Radeon HD 6870 is unlikely to be let down by a lack of memory bandwidth.

The Barts die has a queer rectangular shape and is much smaller than the RV870. Like in every other solution from AMD, there is no heat-spreading cap in the GPU design. The metallic plate on the die package is the only protection. For the first time in the history of the Radeon family the die lacks an ATI logo. There is an AMD logo instead because AMD has made a decision (a very questionable decision, in our opinion) to give up the ATI brand. The rest of the marking is as incomprehensible as before. The only thing you can learn from it is the manufacturing date of this batch of dies. Here, it is the 36th week of 2010. It means that AMD already had large quantities of Barts cores capable of working at 900 MHz in mid-September.

The GPU-Z utility version 0.4.7 supports the Barts and correctly identifies its configuration, except for the revision number. The lack of a checkmark in the OpenCL checkbox is due to the fact that we used an ordinary version of the AMD Catalyst driver rather than the APP Edition which adds OpenCL support. GPU-Z doesn’t report the number of texture-mapping units but we can tell you that the Radeon HD 6870 has 56 of them. Another popular utility, MSI Afterburner, can identify the new Radeon HD cards as well, but its 2.0.0 version cannot control their GPU voltage. The diagnostics panel makes it clear that the GPU frequency lowers from 900 to 100 MHz in the power-saving mode while the memory frequency drops to 300 (1200) MHz. This should make the new card highly economical at low loads.

As we’ve said above, the new Radeon HD family features unprecedented connectivity options. Indeed, there are as many as five connectors at the card’s mounting bracket: two DVI-I ports, two mini-DisplayPorts, and one HDMI. Judging by the marking, you can only use the bottom DVI-I to establish an analog connection by means of an appropriate adapter. The DisplayPorts support DP++ mode. That is, they can output DVI signal via a passive adapter. As described in the theoretical part of this review, the Radeon HD 6800 supports any monitor configurations. It has only one CrossFire connector, so it looks like you cannot combine more than two Radeon HD 6800s into a single graphics subsystem. This feature must be reserved for the more advanced Radeon HD 6900 series.

The cooling system hasn’t changed much and is free from revolutionary innovations. The memory chips and power components are cooled by an aluminum plate which has thermal pads for contact. The GPU is cooled by an aluminum heatsink with a copper base.

The heatsink is rather small but has as many as three heat pipes, two of which are 8 millimeters in diameter. It is fastened to the PCB with four spring-loaded screws and an X-shaped stiff back-plate. A layer of dark-gray thermal grease is applied in the place of contact with the GPU. As you can see in the photo, the cooler’s casing is designed in such a way as to drive some of the fan’s airflow towards the side panel of the system case because the vent grid in the card’s mounting plate is not large. We wouldn’t say that this cooler makes a strong impression, but the Barts is simpler than the Cypress and is supposed to dissipate less heat. So, it should be satisfied with that cooler notwithstanding the increased voltage. The noise factor is the only issue left to be discussed.

Radeon HD 6850: PCB Design and Cooling System

The junior model of the new series is somewhat shorter than its cousin but its power connector is located at the shorter rather than longer edge. Thus, the Radeon HD 6870 and 6850 will have roughly the same dimensions with the power cable attached. The cooler’s casing is designed in the same angular style.

There is nothing interesting we can see until we remove the cooler. Like the senior model, the Radeon HD 6850 has only one CrossFire connector.

As opposed to the Radeon HD 6870, this card has a traditional component layout with the power subsystem residing at the back. Despite the lower clock rate and voltage of the GPU, the GPU voltage regulator has four phases, too.

Like on the senior model, a CHL8214 controller from CHiL Semiconductor is responsible for managing the GPU voltage regulator.

The memory voltage regulator is the same as well, based on an uP6122 chip. These components are located at the front of the PCB. The Radeon HD 6850 has only one external power connector, which is 6-pin, too. Thus, it is going to load the power section of the PCI Express slot more than the Radeon HD 6870, but it has a lower GPU voltage in 3D mode: 1.05 instead of 1.175 volts. The PCB design doesn’t provide for an 8-pin power connector.

We’ve got the same memory chips here as on the Radeon HD 6870: Hynix H5GQ1H24AFR-T2C with a rated frequency of 1250 (5000) MHz. Such chips are rather redundant for the Radeon HD 6850 because this card’s standard memory frequency is only 1000 (4000) MHz. Coupled with the 256-bit memory bus, this ensures a memory bandwidth of 128 GBps. The total memory amount is 1024 megabytes. The card drops its memory clock rate to 300 (900) MHz in power-saving mode.

The GPU marking looks different than that of our Radeon HD 6870. The last line is printed in a different font while the first line, which denotes the manufacturing date, has a letter U. We don’t have a smallest inkling as to what this might mean, but we know that this Barts was manufactured a week later than the Barts installed on our Radeon HD 6870.

The GPU configuration is identified correctly by our software tools. We can only add that the Radeon HD 6850 has 48 active TMUs out of the physically present 56. Like with the other card, MSI Afterburner cannot control the GPU voltage but shows that the power-saving technologies work correctly: the GPU clock rate is lowered to 100 MHz and the memory clock rate to 300 (900) MHz when the card is idle. The graphic core of the Radeon HD 6850 doesn’t have to work at high frequencies, so its voltage is set at 1.05 volts by default.

The junior Radeon HD 6800 series card has the same interfaces as the senior model: two DVI-I and two DisplayPort connectors with support for DP++ and daisy-chain connection. It also has a version 1.4a HDMI port. The card’s single CrossFire connector allows combining two Radeon HD 6850 cards into a single multi-GPU subsystem. Asymmetric configurations with Radeon HD 6870s may be supported as well.

The cooling system of the Radeon HD 6850 is similar to the above-described cooler of the Radeon HD 6870 but simpler: the heatsink is smaller and has only one flat U-shaped heat pipe in its base. The size of the heatsink is far from impressive really. Like on the Radeon HD 6870, the cooler’s casing is shaped in such a way as to drive some of the air towards the side panel of the computer case.

A figured plate with low fins is an additional element of the cooling system. It cools the memory chips and power components, contacting with them via thermal pads. The plate is fastened to the PCB separately from the heatsink and plastic casing.

This cooling system overall doesn’t look capable of performing wonders, either, especially as it has a smaller fan, but the GPU of the Radeon HD 6850 has lower voltage and frequency settings than the same GPU on the Radeon HD 6870. We’ll check out the efficiency of these coolers in the next section of our review.

Power Consumption, Temperature, Noise

It is interesting and important to know how much power your computer components need and consume. Therefore we pay much attention to this issue in our reviews. The new Radeon HD 6 series cards were tested on our standard power-measuring testbed with the following configuration:

The new testbed for measuring electric characteristics of graphics cards uses a card designed by one of our engineers, Oleg Artamonov, and described in his article called PC Power Consumption: How Many Watts Do We Need?. As usual, we used the following benchmarks to load the graphics accelerators:

Except for the maximum load simulation with OCCT, we measured power consumption in each mode for 60 seconds. We limit the run time of OCCT: GPU to 10 seconds to avoid overloading the graphics card's power circuitry. Here are the obtained results:

As expected, the Radeon HD 6870 is much more economical than the Radeon HD 5870, yet its increased GPU voltage shows up in 3D mode where it needs about as much power as the Radeon HD 5850. The new card is far more economical in the low-load modes, though. The load on the +3.3V line is surprisingly high as it hasn’t been used much in recent graphics cards. Otherwise, the Radeon HD 6870 behaves predictably in this test. We had anticipated it to load both its power connectors to the same extent. Indeed, the difference between them is negligible.

The Radeon HD 6850 behaves in a different way. It would invariably produce a result of 30-33 watts in 2D mode although the GPU clock rate, according to MSI Afterburner, was reduced to 100 MHz. We suspect our presale sample of the card didn't have the PowerPlay technology working correctly. For example, the high power consumption at low load may have been due to PowerPlay not reducing the GPU voltage. The same goes for medium loads like HD video decoding: the card proves to need more power than the Radeon HD 6870.

It is in the 3D mode, when the GPU voltage is at its maximum, that we get correct numbers. The Radeon HD 6850 needs much less power than its cousin in 3D applications, which is just what we could expect considering its lower GPU clock rate and voltage as well as the fewer number of active GPU subunits. The Radeon HD 6850 loads the different power lines just like the 6870 model, but it has only one external power connector which is loaded more as the result. This connector has a load of 80 watts in the synthetic OCCT test as the result.

Thus, the new Radeon HD series proves to be very energy efficient in our tests except for the disappointing failure of the PowerPlay technology in the Radeon HD 6850 in some modes, but we don’t think that off-the-shelf versions of the card will have that defect. The junior model of the new series needs but slightly more power than the slower Radeon HD 5770 in 3D applications whereas the senior model is as economical as the Radeon HD 5850 while being faster in modern games as promised by AMD. That’s very good, especially as the Nvidia GeForce GTX 460 1GB is not so energy efficient.

The new Radeon HD models are quite hot at work which is largely due to their not very advanced cooling systems. We must admit that most reference coolers for top-end cards deliver but mediocre performance whereas nonstandard cooling systems are generally much better. So, even though the reference Radeon HD 6870 and 6850 are far from cool, we are sure the graphics card vendors will offer them with better coolers. Besides, the temperature of 75-80°C is quite normal for a modern GPU. It shouldn't worry you at all.

As for the noise factor, the new Radeon HD 6800 cards are very quiet, nearly silent, at low loads, merging with the noise of the other components (the background noise level is 39 dBA in our test lab). But when we launch a heavy 3D application, these cards accelerate their fans and become distinctly audible. The junior model is somewhat quieter than the senior one, according to our noise level meter, but the difference can hardly be caught by ear. We wouldn’t say that these cards are loud. After all, every modern top-end gaming card is noisy. But you must be aware that you won’t have a silent computer if you purchase a Radeon HD 6870 or 6850 unless you choose a special version with a non-reference cooler.

Radeon HD 6800: Multimedia Functionality and HD Video Playback

The UVD video engine has been traditionally improved with each new Radeon HD generation, indicating that the Radeon HD 6800 is positioned by its developer as a solution for HD video playback. Let’s see how good the Barts is in multimedia tasks.

UVD 3.0 offers hardware acceleration for decoding video in a number of formats (DivX/XviD, MPEG2-HD, MPEG4-AVC, MPEG4-MVC, WMV-HD, VC-1, Adobe Flash 10.1, etc.) and supports bitstreaming of many audio formats via HDMI. It can also do hardware post-processing of SD and HD video. In other words, UVD 3.0 is not much different from the previous video engine.

It may look odd that the developer implements hardware decoding support for DivX/XviD and entropy decoding support for MPEG2 in 2010, but UVD 3.0 has been developed not only for graphics cards with a peak power consumption of over 100 watts but also for integration into various mobile GPUs or CPUs. When decoding video, UVD 3.0 must consume less power than a higher-performance CPU, so it is rather a surprise that the Radeon HD 6850 needs about 40 watts of power for playing HD video. This is not much for a desktop computer but quite a lot for a mobile platform.

A desktop PC user doesn’t care much about power consumption. He needs the cooling system to be as quiet as possible and the reference Radeon HD 6850 is not a very quiet card, but the quality of video playback, both in HD formats and in SD formats upscaled to 1080p, is important, too.

In this section of our review we will see how good the Radeon HD 6850 and its UVD 3.0 engine are at decoding Blu-ray, playing HD videos, and upconverting SD video to full-HD.

Video Playback Benchmarking Testbed and Methods

We are going to investigate the decoding performance and playback quality of Nvidia GeForce GTS 450 and other today’s testing participants on the following platform:

The following graphics cards and integrated graphics processors took part in our tests:

We used the following tools to estimate the video playback quality in standard (SD) and high-definition (HD) resolutions:

The driver settings remained the same. However, according to the HQV suite requirements, the noise suppression and detail levels in the drivers were increased to medium (50%-60%), which, however, didn’t affect the results in multi-cadence tests.

Since the owners of high-end sound systems will be extremely interested in the results of lossless threads playback, we also enabled DTS-HD Master Audio and Dolby Digital TrueHD (where available) in order to increase the CPU load in all played movie fragments.

Keeping in mind that all tests are run under Windows 7 OS without disabling background services, the CPU utilization peaks shouldn’t be regarded as critical. It is much more important how much time it takes the CPU on average to complete the task. Note that 1%-2% difference is not indicative of any advantage of a certain graphics accelerator over the competitor.

To estimate the CPU utilization during full-HD video playback (1920x1080) and full-HD video with enabled “picture-in-picture” (PiP) or Bonus View (according to Blu-ray disc Association classification) feature, we used the following movies:

We didn’t use any free content for this test session.

Video Playback Quality

The HQV 2.0 test suite is a means to subjectively evaluate the quality of some video processing operations performed by a graphics card. As we wrote in our earlier reviews, this suite is very detailed and focuses on comparing Blu-ray and DVD players, which are based on specialized video processors. Therefore, today's GPUs do not always score the highest marks in it.

HQV 2.0 DVD

Today, few people watch regular DVD movies on TVs and monitors at the native resolution of DVD content. Most users instead prefer larger screens with full-HD resolution (1920x1080). So, the primary goal of any video processor is not just to properly display video content, but to be able to upscale the image, perform movement correction, reduce noise, improve detail quality, etc. Video fragments used in the HQV 2.0 DVD tests are selected in such a way as to demonstrate how good today’s video processors are at performing each of the mentioned operations.

Announcing its UVD 3.0, AMD did not say anything about image quality improvements. Indeed, the upconverting quality of the Radeon HD 6850 is the same as that of its predecessors.

HQV 2.0 Blu-ray

Similar to HQV 2.0 DVD, the HQV 2.0 Blu-ray test suite allows to subjectively evaluate a video processor at high display resolutions.

Like in the previous case, we don’t see any difference from the older cards’ results. That’s just fine, actually. The Radeon HD 5000/6800 cards deliver higher quality than the competing GeForce solutions and most of their failures (the tests where they score 0 points) occur with noncommercial content. We don't think that users who watch HD movies directly from Blu-ray discs will be disappointed with them. You can only have poor quality when trying to upscale HD content from iTunes and other such services.

With its Radeon HD 6850 series and Catalyst 10.10 driver AMD has begun to set its noise removal and edge enhancement settings at a rather aggressive level by default. We don’t know the reason for that but that makes the new cards score maximum points in the appropriate HQV 2.0 test clips. Unfortunately, the user-defined noise removal setting is far from perfect, making the image fuzzy even at 50%. As a result, many 720p video clips start to look as VHS video cassettes.

Considering that real movies contain numerous scenes shot at different locations with different lighting and often with different cameras, a good video processor must be able to adapt itself to the specific scene on the fly. Therefore we would recommend the user to check out the default video processor settings and adjust them if needed.

By the way, the HQV 2.0 Blu-ray test did not run on the Radeon HD 6850 until we updated it to its latest version but all movies were reproduced excellently. CyberLink PowerDVD 10 with support for AMD Radeon HD 6800 and Blu-ray 3D will be released on November 5, 2010.

When analyzing the results of the HQV tests, you must keep it in mind that the scoring method is highly subjective. Therefore a small difference in the total scores of different graphics cards should rather be neglected.

Blu-ray Playback

Now let’s see how the Radeon HD 6800 series can offload the CPU when decoding HD video.

We don’t see any changes over the previous generation Radeons when playing Dark Knight or Constantine. The new card is good, but not perfect.

The average CPU load when playing our MPEG4-AVC movies on the Radeon HD 6850 is quite low at 7%. The peak load is lower as well, which means you are unlikely to experience sudden slowdowns when watching such movies.

Judging by these results, the entropy decoding of MPEG2 HD on the GPU helps considerably lower the average and peak CPU load. The HD 6850 is the best Radeon in this test.

Summary

Like most of its predecessors, the Radeon HD 6850 is a good choice for a home multimedia computer.

Offering hardware support for decoding video in many formats (including DivX/XviD, MPEG2-HD, MPEG4-AVC, MPEG4-MVC, WMV-HD, VC-1, Adobe Flash 10.1), being able to bitsream all popular types of audio formats via HDMI 1.4a, and featuring high-quality hardware post-processing of SD and HD video, the Radeon HD 6850 is the most advanced graphics card on the market in terms of multimedia capabilities. The downside is that it consumes quite a lot of power and is rather large, so we don’t expect this card to come with passive cooling. As for the HD 6870, it is too long to fit into an HTPC system case.

The Radeon HD 6850 offers higher-quality Blu-ray playback and DVD upscaling than same-class competitors, but still cannot score the maximum amount of points in HQV 2.0. AMD will have to optimize the video engine or driver to achieve better performance in that benchmark.

A special feature of the new card, AMD HD3D technology can output Blu-ray 3D movies to a wide range of TV-sets and projectors without your having to purchase additional software (except for a Blu-ray 3D player like CyberLink PowerDVD Deluxe). Nvidia's competing technology 3D Vision requires that you purchase a special driver/middleware from Nvidia.

Testbed and Methods

We are going to investigate the gaming performance of Radeon HD 6800 products using the following universal testbed:

We used the following ATI Catalyst and Nvidia GeForce drivers:

The ATI Catalyst and Nvidia GeForce graphics card drivers were configured in the following way:

ATI Catalyst:

Nvidia GeForce:

Below is the list of games and test applications we used during this test session:

First-Person 3D Shooters

Third-Person 3D Shooters

RPG

Simulators

Strategies

Semi-synthetic and synthetic Benchmarks

We selected the highest possible level of detail in each game. If the application supported tessellation, we enabled it for the test session.

For settings adjustment, we used standard tools provided by the game itself from the gaming menu. The games configuration files weren’t modified in any way, because the ordinary user doesn’t have to know how to do it. We ran our tests in the following resolutions: 1600x900, 1920x1080 and 2560x1600. Unless stated otherwise, everywhere, where it was possible we added MSAA 4x antialiasing to the standard anisotropic filtering 16x. We enabled antialiasing from the game’s menu. If this was not possible, we forced them using the appropriate driver settings of ATI Catalyst and Nvidia GeForce drivers.

Besides Radeon HD 6870 and Radeon HD 6850, we also tested the following solutions:

Performance was measured with the games’ own tools and the original demos were recorded if possible. We measured not only the average speed, but also the minimum speed of the cards where possible. Otherwise, the performance was measured manually with Fraps utility version 3.2.3. In the latter case we ran the test three times and took the average of the three for the performance charts.

Performance in First-Person 3D Shooters

Aliens vs. Predator

The improve tessellation unit shows good performance. Of course, the new Radeon HD 6800 cards, priced at $179-239, cannot match the GeForce GTX 470 ($259) but the senior model is as good as the GeForce GTX 460 1GB and even beats the latter at 1920x1080 and higher resolutions in terms of bottom speed. We must confess that the gameplay is only comfortable at 1600x900, though. Thanks to the architectural improvements even the Radeon HD 6850 is ahead of the Radeon HD 5870 in this game. And that’s just the beginning.

Battlefield: Bad Company 2

The results confirm AMD’s promise. Having fewer functional subunits, the Radeon HD 6870 competes with the Radeon HD 5850 successfully, even though this is largely due to the difference in their GPU clock rates. The Radeon HD 6850, on its part, does even better than promised, beating the GeForce GTX 460 768MB and matching the GeForce GTX 460 1GB. Considering its lower price, the Radeon HD 6850 looks most appealing but what will the other tests show us?

Call of Duty: Modern Warfare 2

AMD’s promise to deliver the performance of a Radeon HD 5850 is only fulfilled by the Radeon HD 6870 at 1600x900. Starting from 1920x1080, the new card is somewhat slower than its predecessor. Considering the different price ranges, we doubt anyone would want to exchange his Radeon HD 5850 for a Radeon HD 6850, especially as there are still very few games with tessellation. We can note that the 6800 series can be slower than the 5800 cards, though.

Crysis Warhead

Despite being known as a very heavy 3D application, this game doesn’t use tessellation and the Barts cannot show its best here. As a result, the senior model of the new series is only as good as the Radeon HD 5850 while the junior model competes with the GeForce GTX 460 1GB at high resolutions. This is, however, of little practical worth since the frame rates of these cards are only high enough for normal play at 1600x900.

Far Cry 2

Despite its 900MHz clock rate, the Radeon HD 6870 falls farther behind the Radeon HD 5850 as the resolution grows up. The difference amounts to 7% at 2560x1600, which may indicate a lack of memory bandwidth. Fortunately, this only refers to the average frame rates whereas the bottom speeds are roughly the same. Moreover, each card makes the game perfectly playable. The Radeon HD 6850 can only compete with the cheaper GeForce GTX 460 768MB and isn’t very successful in doing that at the resolution of 1600x900. On the other hand, the game runs fast enough at 2560x1600 even on the junior Radeon HD 6800 series card.

Metro 2033

This game is tested without full-screen antialiasing but with tessellation turned on.

The high system requirements of Metro 2033 with tessellation turned on are quite apparent. The GeForce GTX 470 is the only card to deliver more than 40 fps at 1600x900, its bottom speed being as uncomfortably low as 12 fps. The Radeon HD 6870 is a mere 1-3 fps better than the Radeon HD 5850 in terms of bottom speed, which too small an advantage to make any conclusions about the new tessellation unit or other optimizations in the Barts core.

We can again state that the Radeon HD 6800 is slower than the Radeon HD 5800 series.

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

We use the game’s DirectX 10.1 and DirectX 11 modes for graphics cards that support them.

The new cards are more or less competitive to the older Radeon HD 5000 series in this shooter. Tessellation isn’t used extensively in S.T.A.L.K.E.R.: Call of Pripyat, so the new GPUs can’t show their best. As a result, the larger computing resources of the Radeon HD 5800 series make up for the high frequencies of the Radeon HD 6800 cards.

The Radeon HD 6870 is as fast as the GeForce GTX 460 1GB which is officially priced lower by $40. The junior model of the new series looks good, delivering the same speed as the GeForce TX 460 768MB.

Performance in Third-Person 3D Shooters

Just Cause 2

This game’s integrated benchmark does not report the bottom frame rate.

Just Cause 2 doesn’t use tessellation, but has the option of water simulation on the GPU. The Radeon HD 6870 has a GPU clock rate of 900 MHz, which has a positive effect on the speed of geometry processing. Even if the Barts core is only improved in terms of tessellation, the difference in clock rate alone is enough to make this game run at the same speed as on the Radeon HD 5870. That’s a brilliant achievement considering the difference in price between the Radeon HD 6870 and 5870. The Radeon HD 6850 feels good here, too. It equals the GeForce GTX 460 768MB at the two lowest resolutions and allows playing comfortably at 1600x900.

Lost Planet 2

The Barts’ higher tessellation performance is obvious in this game: the Radeon HD 6870 even beats the Radeon HD 5870 in terms of bottom speed at 1600x900. On the other hand, its 22 fps is only a miracle compared to the Radeon HD 5850 but not to the GeForce GTX 460 768MB which easily delivers the same speed. The 1GB version of the latter card maintains a bottom speed of nearly 30 fps, which is unachievable with the Radeon HD 6800 series.

Performance in RPGs

Mass Effect 2

We enforced full-screen antialiasing using the method described in our special Mass Effect 2 review.

Each Radeon HD 6800 card boasts impressive performance, especially at 2560x1600 where they are the only cards besides the more expensive ($259) and hotter GeForce GTX 470 to deliver a playable bottom speed. The Radeon HD 5800 series is slower despite its superiority to the Radeon HD 6800 in a number of technical parameters. They cannot always keep the frame rate above 25 fps.

Performance in Simulators

Colin McRae: Dirt 2

We enable the DirectX 11 mode for graphics cards that support it.

Despite the new tessellation subunit, the Radeon HD 6800 is not as brilliant here as in the earlier tests just because tessellation is not a bottleneck in this game. The senior Radeon HD 6800 is competitive to the Radeon HD 5850 rather than to the Radeon HD 5870. The junior is slower than both GeForce GTX 460 cards except for the resolution of 2560x1600 where it equals the 768MB version. On the other hand, the Radeon HD 6850 makes the game quite playable even at the highest resolution.

Tom Clancy’s H.A.W.X.

The game’s integrated benchmark cannot report the bottom frame rate. We use DirectX 10 and 10.1 modes here.

The new Radeon HD models prove that they indeed belong to a new generation. The Radeon HD 6870 overtakes the GeForce GTX 460 1GB at 1920x1080 and the GeForce GTX 470 at 2560x1600 although this test used to be Nvidia's home turf. The Radeon HD 6850 is not that successful, but quite competitive to the Nvidia GF104-based products staring from 1920x1080.

Tom Clancy’s H.A.W.X. 2 Preview Benchmark

This benchmark was distributed by Nvidia until October 22, 2010.

The Tom Clancy’s H.A.W.X. 2 Preview Benchmark uses tessellation to render the ground, increasing the number of primitives to 1.5 million per one frame, not counting airplanes, trees and buildings in. The size of a typical primitive is 6 pixels, which is not optimal from many points of view.

The preview benchmark (the game itself hasn't been released yet) makes Nvidia’s solutions unrivalled. The Radeon HD 6870 is quite far ahead of the Radeon HD 5870, but cannot match the GeForce GTX 460 768MB, let alone the faster Fermi series products, despite the improved tessellation subunit. The only piece of good news for AMD is that the new cards deliver a playable frame rate even at a resolution of 2560x1600.

We must admit that AMD criticizes this preview benchmark for differing from other tessellation-employing applications. One Web source quotes AMD as saying 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 H.A.W.X. 2, with tessellation on, and that of other games/benchmarks will demonstrate how unrepresentative H.A.W.X. 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 annoyance can be understood because the H.A.W.X. 2 Preview Benchmark uses tessellation too much, making it the main bottleneck in terms of frame rate. It’s interesting to note that the preview benchmark runs faster than the real game H.A.W.X.

Performance in Strategies

BattleForge

We use DirectX 11 mode for graphics cards that support it.

Alas, the Radeon HD series has a very low bottom speed in this game even when we take the new Barts-based solutions. Although the average performance of the Radeon HD 6870 and 6850 is high enough, their bottom speed is downright poor. Meanwhile, the GeForce GTX 460 768MB is quite capable of keeping the frame rate as high as 30 fps even at 1600x900.

StarCraft II: Wings of Liberty

The main achievement of the Radeon HD 6800 in this test is that the bottom speed is much higher compared to the Radeon HD 5850. The senior model of the new series even outperfroms the GeForce GTX 470 at 1920x1080. The resolution of 2560x1600 is, unfortunately, not playable due to the low bottom speed even though the Radeon HD 6870 is close to delivering the desired 25 fps.

Performance in Synthetic and Semi-Synthetic Benchmarks

Futuremark 3DMark Vantage

We minimize the CPU’s influence by using the Extreme profile (1920x1200, 4x FSAA and anisotropic filtering). We also publish the results of the individual tests across all resolutions.

The Radeon HD 6870 scores more than 8000 points overall, surpassing the score of the GeForce GTX 470. The Radeon HD 6850 is in between the two versions of GeForce GTX 460.

The Radeon HD 6800 series is much better in the second test than in the first one, especially the senior model. That test needs high geometrical performance, so this might have been expected. However, we already know from the gaming tests that this is not enough to beat the opponents from the GeForce team.

Final Fantasy XIV Official Benchmark

This benchmark produces an incomprehensible result in points, so we use Fraps to get more understandable numbers. It can only run at 1280x720 and 1920x1080.

We don’t see anything new in this test. It still prefers the Radeon architecture. We can only note that the Radeon HD 6870 is as good as the Radeon HD 5870 at 1920x1080.

Unigine Heaven Benchmark

We use Normal tessellation in this test.

Despite the improved tessellation subunit, the Radeon HD 6800 family doesn’t improve the performance of AMD’s solutions in this benchmark, except for 1920x1080 where the senior model is ahead of the Radeon HD 5870 in terms of bottom speed. Does it mean the Barts is insufficiently effective at complex tessellation or there are some other performance-limiting factors? Whatever the reason, the Radeon HD 6800 series isn’t brilliant in this test.

Conclusion

Having tested the new Radeon HD 6800 series in 19 gaming and synthetic benchmarks, we can make some observations about its performance. The senior Radeon HD 6870 model is a success overall. It is generally faster than the more expensive Radeon HD 5850 and also brings about a number of improvements such as higher tessellation performance which was apparent in a few tests. The summary diagrams make it clear enough.

It must be noted that the new card could not beat the GeForce GTX 460 1GB at 1600x900 but went ahead at 1920x1200 and enjoyed an average advantage of 16% at 2560x1600. Moreover, the Radeon HD 6870 was generally faster than the Radeon HD 5850, the gap being quite large in a few tests. This seems to be a death sentence for the latter card, just as AMD has planned. However, considering the price of the Radeon HD 6870, users who need an affordable but high-performance card for modern games may want to take a look at the GeForce GTX 460 1GB, especially at its factory-overclocked versions with a GPU clock rate of 750-800 MHz. Such cards will be just as good as the Radeon HD 6870 and will also provide such exclusive technologies as PhysX for certain games. Owners of Radeon HD 5870s shouldn’t worry about an upgrade, either, at least until Radeon HD 6900.

The Radeon HD 6850 is an average 15% slower than its senior cousin, the gap amounting to 20-40% in certain games. This new card has no chance against the Radeon HD 5850. It may be faster in tessellation-using applications but such games are rare as yet. It also doesn’t look good against the GeForce GTX 460 768MB. Let’s take a look at the summary diagrams.

The Nvidia card is definitely faster at low resolutions. The Radeon HD 6850 wins but a few tests, and by a very small margin. The cards become equal to each other at higher resolutions. The outcome of their fight depends on the specific game at 1920x1080 whereas the 2560x1600 mode is not meant for graphics cards of this class. Should you upgrade from a Radeon HD 5830 to a Radeon HD 6850? We guess, yes. The newer solution is far more balanced in terms of functionality and performance. But it is not definitely better than the GeForce GTX 460 768MB.

The Radeon HD 6800 models both look competitive in terms of pricing, specs and performance. AMD has done a good job on getting rid of one of the bottleneck in the Radeon HD 5800 architecture, the low speed of tessellation and the low speed of geometry processing at large. Besides, a few multimedia related innovations make these cards unique. We mean their support for DisplayPort 1.2 and HDMI 1.4a, their new video engine that offers hardware DivX decoding, and the opportunity to connect up to six monitors simultaneously in nearly any configuration imaginable.

Considering the power consumption and size of the new cards, we cannot recommend them for typical HTPCs, but the 6850 model might be good enough for a gaming HTPC.

The Barts GPU supports all HD video formats including Blu-ray 3D. It delivers very high, even though not perfect, quality of Blu-ray playback and DVD upscaling according to the HQV 2.0 benchmark. The support for HDMI 1.4a and DisplayPort 2.0 means that the new cards can be used together with modern TV-sets and projects to enjoy stereoscopic 3D content.

So it looks like Nvidia, which was late in releasing its DirectX 11 architecture and transferring all its product lines to it, has been given no time to rest. AMD has used its opponent’s delay to prepare a new and sensitive blow. We are now looking forward to the Radeon HD 6900 “Cayman” to see if it will bring AMD the title of the maker of fastest single-chip graphics cards.

Radeon HD 6870 Summary

Highs:

Lows:

Radeon HD 6850 Summary

Highs:

Lows: