1 year and 1 month after the launch of the Radeon HD 5800 series, the Radeon HD 6800 series is unleashed upon the world. This new line of graphics cards from AMD consists of the Radeon HD 6870, and 6850. Retail prices for these cards are $239 and $179, respectively. However, despite the names, and as their prices imply, these cards are not intended to replace the Radeon HD 5870 and 5850. The replacements for these cards will come at a later date and shall be reviewed at that time. This marks a change in both nomenclature and product positioning for AMD, which will be explained later in this article.
Powering the 6800 series is the new Barts core, an Application Specific Integrated Circuit (ASIC) descended from the Evergreen family of processors which powered the entire Radeon HD 5000 line, but with some new tricks. This Barts core belongs to the Northern Island family of Graphics Processing Units (GPUs). Among those new tricks are improved tessellation performance, thanks to a new tessellation unit dubbed “Generation 7”. Filtering quality is also improved, and a new post-process Anti-Aliasing (AA) mode is introduced dubbed Morphological AA (MLAA). This new method of AA is designed to be compatible with all DirectX applications, which allows gamers to enable AA in all their games. Also new for the 6800 series is stereoscopic 3d support dubbed “HD3D”. This applies to both 3d games and 3d Blu-ray movies. 3d game support has been enabled for 5000 series cards also, but 3d Blu-ray support is exclusive to the 6800 series at this point in time. We do not have any stereoscopic 3d hardware to test so no investigation of this new feature has yet been made. Video decode gets a nice boost in the form of a new revision of Unified Video Decode (UVD) 3. New for this version is Entropy decoding for various popular video formats, as well as Blu-ray support.
Reviewer’s note: this is our first review here @ SemiAccurate. Due to time limitations I have only tested the 6800 series cards, no competing or preceding products. Also, a single testbed was used for all tests and no overclocking was performed. An i7 970 was used in an attempt to minimize CPU limitations, but no guarantee is given that CPU limitations are not present. Built-in “timedemo” style game benchmarks have been used. In future reviews it is our intent to use custom timedemos of actual gameplay. For benchmarks which do not record minimum FPS, subjective “smoothness” is reported. Regarding smoothness, it is this reviewer’s opinion that most games are playable as long as FPS remain above 30 for the majority of gameplay. Preferably, FPS should be a constant 60 or higher to maintain optimum fluidity.
I mentioned previously that Barts is a new ASIC from AMD, descended from the previous generation Evergreen architecture. Barts has been binned to support two SKUs, the Radeon HD 6870 and 6850, internally dubbed “XT” and “Pro”. Physically, these chips are identical. The difference lies in the functional unit count and clock speeds. Barts has the same 5-wide Very Long Instruction Word (VLIW) Arithmetic Logic Unit (ALU) organization as its predecessors. Pre-launch speculation included a reorganization of ALUs ditching the 5th processor known as the “T” unit, however Barts has the same ALU organization as its predecessors. Barts also features the previously-mentioned 7th generation tessellation unit which dramatically improves tessellation performance thanks to changes in thread management and buffering. If you’d like to know more about the changes Barts brings to the table you can read about them in our Barts Architecture Article here.
As mentioned previously, both 6800 series SKUs feature the same Barts GPU, but they are binned to create differences that allow AMD to sell more than one product based upon Barts. Physically, Barts is approximately 255mm square and is comprised of some 1.7 billion transistors. Above we see just what makes the Radeon 6870 different from the 6850. A healthy clock increase is built in to every 6870, as well as an additional two Single Instruction Multiple Data engines (SIMD) which accounts for an ALU increase of 160. Accompanying those SIMDs are two more quad-texture units (TMUs). A 5% memory clock increase over the 6850 is also featured in the 6870.
The largest difference between the 6870 and the 6850 comes in the form of its texturing and math capabilities, both clocking in at just over 35% thanks to the additional functional units and the increased clock speed of the 6870. The rest of the differences in fill-rates, triangle setup rate, and bandwidth are due purely to their respective clock speed differences, approximately 16% and 5%.
The 6870 is the largest of the cards at 10.5”, 11.0” with the fan shroud. The reference 6850 is much shorter at only 9.0”, and Sapphire’s custom 6850 is even shorter at 8.5”. The cooler does not extend beyond the length of the PCB for either of the 6850s. The 6870 is, unsurprisingly the heaviest of the cards with the reference 6850 weighing slightly less due to its smaller heatsink and shorter PCB dimensions. The Sapphire 6850 is significantly lighter than the reference design despite making use of a much larger cooler. This is due to the reference design’s use of a large metal bracket which cools the RAM chips and the voltage regulators, something which Sapphire chose to do without for their 6850. This did not seem to negatively affect board temperature or stability of the Sapphire card.
Below are all three cards lined up for size comparison purposes. You can clearly see the difference in length between them.
How about some hot naked silicon action? Here are the two 6850s without any clothes on. Ooh yeah, show us your dirty bits, you’re naughty graphics cards aren’t you? Erm, sorry, I might’ve gotten a little carried away there…
On a serious note, you can see that Sapphire chose to relocate the power phases towards the rear of the card, away from the GDDR5 chips. This was likely a wise move on the part of the engineering team at Sapphire since the GDDR5 chips have no heatsinks attached and the power phase components are some of the hottest components on any graphics card.
Below is the 6870 in all its glory.
Much like Sapphire’s custom design,the 6870’s power phase components are located towards the rear of the card, away from the GDDR5 chips.
Here are the heatsinks for the 6850s, Sapphire on top, reference on bottom. Later you’ll see what effect these have on temperature and power consumption.
Here’s another shot of the reference 6850 heatsink without the fan shroud sitting next to an Intel 2.5″ SSD for comparison purposes. Rather small, wouldn’t you say?
Both the 6870 and the 6850 offer plenty of connection options which allow the user to connect multiple monitors simultaneously. The reference design for both the 6870 and the 6850 features two mini Display Port 1.2 (DP) connectors, each of which can support up to three monitors when used in conjunction with a DP 1.2 hub. At the time of this writing however, none of the board partner 6850s on the market are using the mini DP connectors but instead feature a single full-sized DP connector which offers three monitor connectivity in the same manner as the mini-DP connectors. Also common to both 6870 and 6850 are two DVI ports, only one of which is dual-link. Additionally, an HDMI 1.4a port is featured on both cards. The Sapphire 6850 even includes an HDMI cable.
The HD 6800 series brings some image quality (IQ) improvements over the HD 5000 line. Anisotropic filtering (AF) quality has improved, due to an enhanced algorithm designed to reduce noise within textures and smooth transitions between filtering levels. Another major new IQ enhancer is MLAA, a post-process filter that runs through the shader core via DirectCompute kernel. MLAA is purportedly compatible with all DirectX 9/10/11 applications. MLAA first appeared in the console world in the Playstation 3 title Saboteur, and later, in God of War 3. Unfortunately, due to time constraints and some technical issues, screenshots of MLAA in action were not taken. These will be included in a future review.
These new IQ enhancers are accessible in the 3d section of the redesigned control panel in the Catalyst Control Center (CCC). A new slider for Catalyst A.I. filtering quality has been added which allows the user to choose between various AF quality levels, High Quality, Quality, and Performance. The High Quality setting disables all texture optimizations which could cause visual artifacts. Both Quality and Performance settings enable anisotropic and trilinear filtering optimizations designed to improve performance, with little to no degradation of IQ for the Quality setting and “a small impact” on IQ at the Performance setting. These settings are visible below.
Beneath the Cat. A.I. filtering quality slider is a checkbox labeled “Enable Surface Format Optimization”. This feature is designed to improve performance of some textures in games that use 16-bit HDR rendering and is said to have “no discernable effect on image quality.” All testing in this review will be performed with this feature disabled due to some controversy regarding its impact on IQ.
Using 3dcenter.org’s filter tester, screenshots of the three levels of filtering quality now offered have been taken. This is filtering set to “Performance”.
On the left is the TMU rendering implemented by AMD and on the right is the reference ALU rendered implementation. It’s difficult to make out in this screenshot because this image has been resized for viewing in this article. Due to file size, we have put up links to the tests in Performance, Quality, and High Quality modes. all saved as 90% quality jpegs. If you want the full uncompressed versions, please email the editor, (editor at semiaccurate dot com).
With these quality settings if you look at the center of the image you can see that as you step up the quality the texture becomes sharper in the center. Screenshots don’t do this justice, however, you really need to see it in motion to get a sense for the quality. Some texture “shimmering” was observed in both the reference filtering implementation as well as AMD’s TMU filtering implementation during this test, more so on AMD’s TMU implementation than the reference. However, this shimmering becomes less noticeable as you increase the filtering quality preset. At the “HQ” setting, the shimmering produced is only slightly worse than reference. That being said, this is a synthetic test designed to showcase a “worst-case” scenario and no texture shimmering was noticed during benchmarking using the HQ preset.
open test stand
ambient temperature ~ 20 Degrees C
Intel i7 970 3.2GHz (24×133)
Coolermaster Hyper 212 Plus
OCZ Platinum 6GB DDR3 1600 7-7-7-19 @ 1.66V
Enermax Galaxy Evo 1250W
WD 500GB Caviar RE2
AMD Radeon HD 6870 engineering sample
900MHz core 1050MHz memory
AMD Radeon HD 6850 engineering sample
775MHz core 1000MHz memory
Windows 7 Ultimate 64
Radeon Driver 8.782 RC2 (beta)
Catalyst Application Profile 10.9a
Chipset driver 22.214.171.1247 (beta)
AA: application settings, standard filter
AF: application settings
Catalyst A.I.: filtering quality: HQ, surface format optimization: disabled
Vsync: off unless applciation specifies
AA mode: performance multi-sample AA
Our testing methods are simple. A range of synthetic benchmarks are used to highlight the theoretical maximum performance capabilities of each product, whilst “real-world” applications are used to show performance in applications that a reader and potential customer may use with these products. All tests are run 3 times, with the reported results averaged across those 3 runs. All graphical settings are set to their highest levels, unless otherwise noted. This includes IQ enhancers such as AA and AF.
3dmark Vantage is a synthetic benchmark which tests graphics card performance in theoretical game scenarios, as well as specific aspects of graphics rendering including pixel and texture fill-rates, and various shader throughputs. It is a DirectX10 benchmark. A DX11 successor tentatively titled “3dmark 11” is in the works.
The GPU score is the most important number here as it more accurately reflects the performance of the graphics card(s) in question. Some very interesting scaling is observed in this benchmark. The 6870 is only 22% faster than the 6850 at the “performance” setting, a less graphics intensive setting which runs at a mere 1280×1024 screen resolution and has overall low to medium graphics settings. Moving on to the “high” setting, we see greater performance scaling in the 6870’s favor at 35%. Lastly, the “extreme” setting produces a whopping 41% performance differential between these two cards. Looking at the following “feature” tests, we can see how these cards handle certain aspects of graphics rendering and achieve better understanding of the performance difference between the two cards.
A brief explanation of the feature tests is in order. Feature Test 1 tests texture fill-rate. The Radeon 6870 has approximately 35% higher texture fillrate than the 6850, which is the precise difference between the cards in this test. Feature Test 2 tests pixel fill-rate. In this test the 6870 tests only about 3% higher than the 6850 but the theoretical performance difference between these cards should be 16%. Also, the tested pixel fill-rate is significantly below the maximum theoretical pixel fill-rate for both cards, indicating some other bottleneck is at work here. Feature Test 3 uses Parallax Occlusion Mapping to test shader throughput. The difference here is nearly 32%, close to the 35% theoretical difference between these cards. Feature Test 4 is a cloth simulation meant to test the geometry shading capabilities of the cards. A 13% difference is measured here. Feature Test 5 is a particle simulation meant to test the vertex shading capability of the cards. A difference of 8% is measured. Finally, Feature Test 6 is a perlin noise test designed primarily to test shader throughput. A 35% difference is observed between the 6870 and the 6850, in line with the theoretical maximum shader throughput differential between these cards.
3dmark 06 is the direct predecessor to 3dmark Vantage. It utilizes DirectX9 to test various aspects of graphics rendering. Until the release of 3dmark Vantage, 3dmark 06 was perhaps the most widely used synthetic graphics benchmark.
Here we have the overall and Shader Model scores for 3dmark 06. The overall scores show little difference due to the low default test resolution of only 1280×1024. However, the Shader Model 3 score shows a nearly 11% difference between the cards in this more modern shader limited test. I use the term “modern” begrudgingly, given the fact that DirectX11 uses Shader Model 5.
Again we see little difference between the cards in these tests, though we now see the individual tests responsible for the SM3 score in the previous graph, the “HDR” tests.
Unigine Heaven 2.0
Heaven is a synthetic benchmark designed to test tessellation performance. Multiple levels of tessellation are selectable. For the purposes of this test, “normal” and “extreme” levels have been selected and tested at a resolution of 1920×1200. Anti-aliasing and anisotropic filtering have also been disabled in order to stress the impact of tessellation.
The overall score for Heaven is shown in the above graph. The “normal” setting has the 6870 outperforming the 6850 by approximately 13%, slightly off the maximum theoretical difference of 16% as determined by the triangle setup rate which itself is governed by the GPU clock speed of the respective cards. The difference between the two cards when tested at the “extreme” setting shifts slightly more toward the theoretical maximum, with the 6870 outperforming the 6850 by approximately 14%. The results are broken down further in the following graph which shows the FPS during these tests.
The overall score given by Heaven is based on the various FPS values produced during the benchmark run. Using the “normal” setting, the 6870 is faster than the 6850 by 15%, 14%, and 10% respectively using minimum, average, and maximum frame-rate values. Using the “extreme” setting, the 6870 is faster than the 6850 by 17%, 15%, and 11% respectively using minimum, average, and maximum frame-rate values. When these values are averaged, they align nicely with the 13% and 14% performance differentials between the overall scores of the two cards.
Aliens vs. Predator
Aliens vs. Predator is a first person shooter title which makes use of DirectX 11. Tessellation is featured in the game’s bag of graphical tricks, as are advanced shadows, Ambient Occlusion (AO), and motion blur. For testing purposes of this title, all graphical options have been set to their highest values. Changes in resolution and AA have also been made for comparison purposes. Resolutions of 1680×1050, 1920×1200, and 2560×1600 have been tested. AA was set to either off, or 4x. AF was set to 16x. Average FPS was recorded, however minimum and maximum were not.
Starting with the 1680×1050 tests, both cards are very capable of playing the game at this resolution with or without AA. At this resolution the 6870 is 28% faster than the 6850 with AA, and 38% faster than the 6850 without AA. Enabling AA at this resolution causes an almost 39% drop in frame rate for the 6870, and close to a 34% drop for the 6850. Moving on to 1920×1200, the average FPS drops but remains playable for both cards. At this resolution the 6870 is 28% faster than the 6850 with AA, and 38% faster than the 6850 without AA. Using AA at this resolution causes the 6870’s FPS to drop by the same 39% as seen at 1680×1050. The 6850 has the same performance loss of 34% as seen at 1680×1050. When AA is enabled, the 6870 is still playable, though the 6850’s FPS drop enough that some may consider running without AA at this resolution. At 2560×1600, the 6870 is about 27% faster than the 6850 with AA, and almost 37% faster than the 6850 without AA. Once again we see the same loss of performance on both cards with AA enabled as observed at lower resolutions, 39% and 34% for the 6870 and 6850, respectively. At this resolution only the 6870 remains playable, and then, only without AA.
Batman Arkham Asylum
Batman Arkham Asylum is a third-person action/adventure title which utilizes the Unreal 3 engine and DirectX 9 rendering. All in-game options have been set to their highest levels with the exception of PhysX since it cannot be accelerated by AMD graphics cards. PhysX can still be enabled in this game should the user choose to do so, however it will run on the CPU which has been demonstrated to dramatically reduce performance when not accelerated via GPU. Tested resolutions were 1680×1050, 1920×1200, and 2560×1600. AA was set to either off, or 4x. AF was set to 16x. Minimum, average, and maximum FPS was recorded.
At the resolution of 1680×1050, both cards are capable of playing this game without any noticeable drop in FPS. Even with 4x AA enabled the game never drops below 61 FPS on the 6850. On average, the 6870 is approximately 15% faster than the 6850 with AA, and nearly 13% faster than the 6850 without AA. With AA enabled on the 6870, performance drops by about 34%. The 6850 loses 35% of its performance with AA.
Moving on to 1920×1200 tests, both cards still remain perfectly capable of playing this game with AA at the highest settings. Minimum FPS on the 6850 with AA enabled is still a whopping 52 FPS, with the 6870 never dipping below 60. On average, the 6870 is almost 15% faster than the 6850 with AA, and 18% faster than the 6850 without AA. Turning on AA at this resolution means you will give up about 37% of your frame-rate on a 6870, and close to 35% on a 6850.
Finally, even at the high resolution of 2560×1600, both cards remain playable. With AA enabled the 6850 never drops below 30 FPS, which is still comfortably playable in my opinion. For those of you that prefer FPS to remain at or near 60 at all times, the 6870 is able to achieve this level of performance when AA is disabled, and the 6850 can maintain FPS above 50 also without AA. On average, the 6870 is 16% faster than the 6850 with AA, and 19% faster than the 6850 without AA. The 6870 loses 38% of its performance when using AA at this resolution, while the 6850 sees a drop in FPS of approximately 36%.
Just Cause 2
Just Cause 2 is a first-person shooter title using a custom DirectX 10 engine dubbed “Avalanche 2.0”, named after the development studio. The game also makes use of Nvidia’s proprietary Compute Unified Device Architecture (CUDA) to create some advanced graphical and physics effects. Most notably, water is simulated via CUDA which produces a more true-to-life implementation, both physically, and graphically. AMD graphics cards do not support CUDA so this option is not available for testing purposes. Tested resolutions were 1680×1050, 1920×1200, and 2560×1600. AA was set to either off, or 4x. AF was set to 16x. Changes to Screen Space Ambient Occlusion (SSAO) and shadow quality have been noted, where appropriate. Minimum, average, and maximum FPS was recorded. In an attempt to maintain playable frame-rates in this game, some settings have been altered when AA is enabled so no comparisons between AA on and off will be made. The “Concrete Jungle” level was tested as it is the most demanding of all the included benchmarks.
Starting off at 1680×1050, both cards are perfectly capable of playing the game at any of the settings tested. The 6870 is on average 20% faster than the 6850 without AA, and a little more than 14% faster with AA.
At 1920×1200 the increased workload is measured to produce a slight decrease in frame rate on both cards, but both remain playable in my opinion. The 6870 is on average 17% faster than the 6850 without AA, and a little less than 17% faster with AA.
2560×1600 proves to be a difficult resolution for these cards, more so for the 6850 than the 6870 which for the most part remains playable. It is recommended to disable SSAO and advanced shadows to maintain playable FPS at this resolution on a single 6850 or 6870. Without AA the 6870 is about 21% faster than the 6850. With AA the 6870 is on average approximately 18% faster than the 6850.
Mafia II is a third-person action/adventure title from developer 2k Czech. It uses the custom Illusion engine and DirectX 9. It also supports PhysX which was not tested. Average FPS has been recorded at the resolutions of 1680×1050, 1920×1200, and 2560×1600. AA on/off results have been recorded also.
At 1680×1050, both cards handle this game with ease, both with and without AA. The 6870 is 18% faster than the 6850 at this resolution, regardless of whether AA is enabled. Using AA results in a drop of nearly 30% in FPS on the 6870, and the same 30% for the 6850.
1920×1200 is still playable on both cards with and without AA. The 6870 is 19% faster than the 6850, again both with and without AA. Both cards lose 32% of their FPS when AA is enabled.
2560×1600 is difficult for the 6850 to pull off, with visible dips in FPS throughout especially when AA is on. The 6870 handles this resolution a bit better, though the use of AA brings the average FPS under 30. When AA is enabled the 6870 is approximately 21% faster than the 6850. Without AA the 6870 is nearly 28% faster. Using AA causes the 6870 to lose nearly 42% of its performance, the largest hit of any game tested thus far. The 6850 sees its frame-rate drop by 39%, also the largest drop observed with AA for this card in any title tested.
Metro 2033 is a first-person shooter title from 4A Games. It uses the custom 4A engine which supports DirectX 11. It also supports PhysX, which was not tested. Numerous advanced DirectX 11 effects are available in this game, including Tessellation, Depth of Field (DOF), Motion Blur, and Ambient Occlusion (AO). The use of these effects are noted where applicable. The resolutions of 1680×1050, 1920×1200, and 2560×1600 are tested. In-game Analytical AA (AAA) was also used, and is noted appropriately. The level “Frontline” was tested as it is one of the more demanding scenes in the game.
At 1680×1050 using DX11 and all settings at their highest levels including in-game AAA, even the 6870 is seen to struggle at times, though I would say it borders on playable. The low minimum FPS number in this game comes at the very beginning of the benchmark when the level is still loading. Actual minimums were much closer to average. If the FPS seems a tad uncomfortable, reducing one or more graphical settings should make the difference. At these settings the 6870 is about 18% faster than the 6850.
Changing from DX11 to DX9 yields a very different playing experience. The graphical effects are less pronounced, though still rather striking in my opinion. Also, performance is good enough that both cards can play the game quite well at these settings. Switching to DX9, the 6870 is a whopping 72% faster than the DX11 codepath. The 6850 is also much faster, 69% in fact. Comparing the two cards at these settings the 6870 is about 20% faster than the 6850.
Changing the settings slider down another notch to “high” keeps most of the major graphical effects of “very high” but loses AAA. This results in another substantive performance boost for both cards. The 6870 is an astounding 150% faster at these settings than when using the highest settings and DX11. It is also 46% faster than when using the DX9 “very high” graphical preset. The 6850 is 136% faster here than when using the highest settings and DX11. It is also 39% faster compared to DX9 “very high”. The 6870 is about 25% faster than the 6850 at these settings.
Moving on to 1920×1200 and dropping down to the “high” settings preset, the game is about as playable as at 1680×1050 very high settings and is a good alternative. The 6870 is about 16% faster than the 6850 here.
If you desire more performance but wish to maintain the 1920×1200 resolution, the best compromise between image quality and performance is the “high” settings preset, though “very high” should also be playable for many people. The 6870 is almost twice as fast at these settings than when using DX11 “high” settings. Unfortunately the scaling with the 6850 isn’t quite as good, though 78% is still impressive. Comparing the two cards at these settings the 6870 is approximately 29% faster than the 6850.
Finally at 2560×1600 using the DX9 codepath the “medium” settings preset is the most playable, though many would find the “high” preset to be acceptable as well. Here the 6870 is about 25% faster than the 6850.
S.T.A.L.K.E.R Call of Pripyat
S.T.A.L.K.E.R. Call of Pripyat is a first-person shooter title from GSC Game World which uses the XRAY engine, a DX11 engine. Among the advanced DX11 graphical effects in this game are tessellation, SSAO, and advanced lighting and shadowing. This game has been tested in DX11 at the resolutions of 1680×1050, 1920×1200, and 2560×1600. AA was also tested both on and off and is noted appropriately.
Starting at 1680×1050 both cards are very playable. The 6870 is 32% faster than the 6850 with AA, and 25% faster without. Enabling AA causes a performance drop of about 30% with the 6870, and 34% with the 6850.
At 1920×1200 the 6870 remains playable with and without AA. The 6850 does fine without AA but when it is turned on the performance suffers enough that some may not be comfortable with the results. Here the 6870 is 30% faster than the 6850 when using AA, and a solid 41% faster than the 6850 when not using AA. Enabling AA results in a performance loss of about 40% for the 6870, and 35% for the 6850.
Finally, 2560×1600 proves to be too much for these cards in most situations when using the highest settings, though the 6870 is playable without AA.
Power, Noise, and Heat
The power consumption of all three cards was astoundingly low, special note going to the Sapphire custom 6850 which is head and shoulders above the reference design. Thanks to its custom cooler, load temperature on the Sapphire 6850 was a remarkable 27 degrees C lower than the reference design. Also noteworthy, the noise difference between the Sapphire 6850 and the reference 6850 was night and day. The reference 6850 at its peak fan speed of 60% was noticeably louder than the other system fans while the Sapphire 6850’s fan was indistinguishable from the other system fans. The reference 6870 with its much larger cooler was also quieter than the reference 6850, and thanks to the low temperature produced by its superior cooler it also consumed only slightly more power under load, a meager 2W.
AMD has filled a large gap in their graphics card product lineup with the introduction of the Radeon HD 6870 and 6850. Given the performance demonstrated by these cards in our testing, their names, and recommended prices, AMD is taking the Radeon family in a new direction. For the most part this is a welcomed change, but some may be put off by the names of these new cards and the performance implications thereof. For what it’s worth my opinion is this is much ado about nothing. It is a break in nomenclature, to be sure, but the 6800 series products are cheaper than their 5800 predecessors so a direct comparison cannot be made without also taking this into consideration.
For gamers looking to play the latest games with high in-game settings, the Radeon 6870 and 6850 offer good value. Both cards have proven capable of playing all of the games tested here at a minimum of 1680×1050 with little difficulty. In most cases, 1920×1200 (or 1080P) gaming is also a viable option. In many cases even 2560×1600 is in reach, albeit sometimes requiring a reduction in settings.
Gaming performance, heat, power, noise, and image quality – these cards have the best in each category. The 6870 costs about 33% more than the 6850 and offers anywhere from 15% to 41% more performance depending on game and settings. The more you increase the settings, the better the return on your investment. If you game at lower resolutions and are on a strict budget of less than $200 though, you won’t go wrong with a 6850 either. Only the most demanding gamers will want more than a single 6870 or 6850 can offer, and for that reason I recommend these cards to anyone searching for a new graphics card in the $179-$239 price bracket. Though no reference 6850 boards are being sold at this point in time, clearly the Sapphire custom design is a cut above, and should reference boards ever appear on the market I would recommend picking up the Sapphire card over a reference board every time.
A Look Ahead
Next up from S|A laboratories is a Crossfire performance expose featuring the same Sapphire 6850 cards and two as yet to be determined 6870s. Overclocking will be explored, among other things.S|A