reviewamds radeon hd 6850 and 6870 graphics processors

AMD’s Radeon HD 6850 and 6870 graphics processors

For a journalist, there’s nothing better than having a good story to tell. At least, that’s always been my way of thinking, and we’ve had no shortage of intrigue, one-upsmanship, and swings of momentum in the GPU arena over the past year or so.

AMD grabbed the lead with the debut of its DirectX 11-class Radeon HD 5000-series graphics processors last September, well ahead of long-time rival Nvidia’s competing chips. These new Radeons were quite good products, with a few strokes of brilliance like the Eyefinity multi-monitor gaming feature, but those highlights were counterbalanced by a frustrating series of supply problems stretching into 2010 caused by TSMC’s troubled 40-nm chip manufacturing process. That same chipmaking process was a major contributor to uncharacteristically long delays in Nvidia’s DX11 GPUs, which left a frustrated AMD with a market largely all to itself—a market it couldn’t fully supply. Consumers groaned as a nearly unprecedented thing happened: prices on Radeon HD 5800-series cards rose above their introductory levels—and held there.

At the very end of the first quarter of the year, the first Fermi-based GeForces finally arrived. They ran hotter and louder but not much faster than the Radeon HD 5870, not exactly a winning combination. The outlook for Nvidia looked rather dim at that point, but a funny thing happened on the way to AMD’s coronation as the kings of the DX11 generation. The new GeForces’ performance quietly crept upward as Nvidia tuned its drivers for this novel, unfamiliar architecture, and then, in the middle of July, the GF104 debuted. This GPU, derived from the Fermi architecture, was smaller and more tightly focused on achieving strong performance in today’s games. Onboard the GeForce GTX 460, it gave the incumbent Radeons much stiffer competition. Soon, we were declaring the GeForce GTX 400 series the new kings of value and hinting strongly that AMD needed to cut Radeon prices to win our recommendation.

Oddly enough, AMD didn’t budge for a while, likely because supply constraints meant the firm was selling all of the graphics chips it could secure from TSMC. But AMD had, well, another card or two up its sleeve that would allow it to challenge the GTX 460 much more directly. Those cards, we now know, are called the Radeon HD 6850 and 6870, a pair of new offerings that come as part of AMD’s annual fall refresh of its GPU lineup. They are both based on a leaner, meaner new graphics chip code-named Barts, a part of AMD’s “Northern Islands” series of GPUs.

Barts? Where’s Homer’s?
The funny thing about Barts is that it’s made using the exact same 40-nm fabrication process that has caused both AMD and Nvidia no end of trouble, mostly because AMD had little choice in the matter when TSMC outright canceled its plans for a 32-nm fabrication process. Both of the major GPU makers had to adjust their plans rather abruptly at that point, focusing on improvements to their chip designs to deliver additional goodness in this next generation of products.

Yet in the midst of some real frustrations, there’s good news on several fronts. AMD Graphics CTO Eric Demers told us last week that TSMC had finally gotten a handle on the problems with its 40-nm process technology over the summer. If so, the latest chips from both AMD and Nvidia should be cheaper, faster, and more plentiful. That trend should be reinforced by some choices AMD has made along the way, especially the fact the Barts is actually smaller—and thus cheaper to produce—than the Cypress chip it replaces. Barts’ mission is to address the value and performance sweet spot in the middle of the market, obviously opposing the GeForce GTX 460. Although the cards based on Barts are dubbed 6850 and 6870 and promise performance fairly similar to the products they replace, they should be less expensive, draw less power, and produce less heat than their predecessors.

A block diagram of the Barts GPU. Source: AMD.

The image above maps out the major components of the Barts chip in a familiar fashion. For the most part, this is the same core GPU architecture we know from the Cypress chip behind the Radeon HD 5800 series, only scaled down slightly and tweaked in several ways. Cypress has 20 SIMD arrays, each with 16 five-ALU-wide execution units, giving it a total of 1600 arithmetic logic units, or ALUs, with which to process the various types of shaders involved in the DX11 graphics pipeline. Barts dials back the SIMD array count slightly to 14, giving it a grand total of 1120 shader ALUs. With this GPU architecture, that change has some natural implications. The texture units, for instance, are aligned with the chip’s SIMD arrays, so those drop in number proportionally, as well. Here are the vitals on Barts and some of its closest friends, to give you a sense of things.


width (bits)
process node


48 64 512 4 384 3000 529* 40 nm

32 64 384 2 256 1950 331* 40 nm

16 40 800 1 256 956 256 55 nm

32 80 1600 1 256 2150 334 40 nm

32 56 1120 1 256 1700 255 40 nm
*Best published estimate; Nvidia doesn’t divulge die sizes

With the GF104, Nvidia held texturing capacity steady at the GF100’s rate while reducing nearly everything else—ROP rate, rasterization rate, memory interface width, and ALU count. The result was a GPU probably better tuned to the needs of current games.

With Barts, AMD has made a different set of choices, reducing shader processing and texturing capacity versus Cypress while retaining the same ROP rate and memory interface size. Oddly enough, these very different choices may also produce a GPU better tuned for the usage patterns of today’s game engines, given the present state of AMD’s GPU architecture. After all, Cypress doubled up on RV770’s resources in nearly every way but memory bandwidth. If that left it, at times, with an excess of shader and texturing power, then Barts may well be a more optimal balance of resources overall. That may especially be the case when high levels of antialiasing are in use, since Barts has the same ROP blending power, clock for clock, as Cypress—and as a smaller, newer chip, Barts may have a little more clock speed headroom.

Cypress (left) versus Barts (right)

By the way, you may have noticed the presence of two “ultra-threaded dispatch processor” blocks in the diagram above, and if you’re into these things, you may have recalled that the diagrams of Cypress only showed one of these blocks. Truth is, though, that this diagram of Barts is simply more detailed than the earlier one of Cypress. AMD’s David Nalasco tells us both chips have dual “macro sequencers,” as AMD calls them internally, to “dispatch instructions to the SIMDs.” (There’s also a “micro sequencer” in each SIMD.) As the diagram shows, each macro sequencer has instruction and constant caches. One bit of detail missing above is a crossbar between the two “rasterizer” blocks and the macro sequencers, so either sequencer can be fed by either rasterizer.

To take you further down the rabbit hole, the presence of two rasterizers in the diagram above may be a little bit misleading. As with Cypress, Barts has dual scan converters, but it lacks the setup and primitive interpolation rates to process more than one triangle per clock cycle. That’s in contrast to the GF104, which can process two polygons per clock tick, or the GF100, whose max is four.

Although the setup rate hasn’t changed in Barts, the chip’s internal geometry processing throughput should be higher thanks to some selective tweaks. One of DirectX 11’s key features is tessellation, in which a relatively low-polygon model is sent to the GPU, and the chip then adds additional detail by using a mathematical description of the surface’s curves and, sometimes, a texture map of its bumps. Adding detail once the model is on the chip can reduce host-to-GPU communications overhead, oftentimes dramatically; it also makes much higher degrees of geometric complexity feasible. One of the challenges tessellation presents is the management of data flow. As essentially a very effective form of compression, tessellation involves a relatively small amount of input data and a much larger, sometimes daunting amount of output data. To better deal with this data flow in Barts, AMD “re-sized some queues and buffers,” according to Nalasco, “to achieve significantly higher peak throughput” in certain cases. At the same time, thread management for domain shaders, which handle post-expansion geometry processing, has been improved.

AMD claims these changes had “negligible impact” on Barts’ transistor budget and power draw, yet the firm has measured tessellation throughput for Barts at up to twice that of Cypress in directed tests. The biggest gains come at lower tessellation levels, as show in the image below. At higher levels, the chips’ common setup rate likely becomes a limiting factor, and the two are separated only by Barts’ slightly higher clock speed.

Barts vs. Cypress tessellation throughput. Source: AMD.

Interestingly enough, we were able to measure a substantial difference between Cypress and Barts ourselves using the hyper-tessellated Unigine Heaven demo.

Barts hasn’t quite matched the GF104 and friends, with their truly parallel geometry processing capabilities, but it has narrowed the gap quite a bit.

Barts also has some image quality improvements, one in hardware and one in software, that we’ll discuss shortly, but that’s about it in terms of changes to the core graphics hardware. We were a little bit surprised to see Demers claiming rather large gains in performance per chip area for Barts versus Cypress, on the order of 25%, given that the two chips share the same underlying architecture and are made on the same fabrication process, but that’s precisely what happened during the press event for this product. Strangely, the comparison being made was between the Radeon HD 6870—a fully enabled Barts chip running at peak clock speeds—and the Radeon HD 5850—a partially disabled Cypress variant with lower clocks. I also run faster than Usain Bolt if you cut off one of his legs below the knee, but that’s not something I like to advertise.

Texture filtering quality improvements
We were pleased when Cypress and the Radeon HD 5000 series introduced revised texture filtering with some nifty properties, including angle-invariant anisotropic filtering. Although that’s just as geeky as it sounds, the real-world impact is noteworthy, because texture filtering has a huge influence on image quality. If objects shimmer, sparkle, or crawl as you move around in a game, yeah, that’s probably poor texture filtering.

In other words, “bad filter make Thog’s Xbox suck.”

We know how to filter textures to eliminate such artifacts, but doing so requires lots of sampling and is, in performance terms, rather expensive. As a result, GPU makers have devised shortcuts, attempting to produce the best compromise between image quality and performance. Some of those filtering algorithms have been pretty complex, and although they haven’t all been great in every way, they’ve allowed us to cope. That’s often the name of the game in real-time graphics.

Over time, as transistor budgets have grown, the trade-offs between performance and quality have become less stark. Cypress represented a high-water mark of sorts because it promised to eliminate one of the worse compromises in older filtering algorithms, the fact that surfaces at some angles of inclination weren’t filtered as well as others, while improving filtering quality overall.

Trouble is, after the Radeon HD 5000 series had been in the market for while, folks started noticing some problems with Cypress’ texture filtering, especially in textures with lots of fine, high-contrast detail. This problem wasn’t evident in every case—heck, I never noticed it myself while gaming on a Cypress card—but it turned out to be quite real. At the press event for Barts, AMD Graphics CTO Eric Demers admitted that it was an issue with Cypress-era hardware.

We’ve replicated an example he showed from the D3D AF Tester application using a high-frequency checkerboard texture. In the image below, you’re looking down a 3D-rendered cylinder with that texture mapped to the interior walls. As the squares in the checkerboard become much too small to represent with a single pixel, the goal of good filtering is to produce a smooth, visually comprehensible representation of what’s happening at a sub-pixel level. On Cypress, the image produced looks like so:


There are several very obvious transition points that form rings within the image above. Those obvious transitions represent a failure of blending, and in a game with, say, a very detailed texture of a road stretching out ahead, they have the potential to translate into visible lines that travel ahead of you, looking cheesy. (For those of us old enough to remember the bad old days of bilinear-only filtering on early 3D chips, this effect might induce flashbacks.)

Demers relayed to us the dismay that he and his team had when they realized this problem made it into Cypress hardware. They thought they’d created a very elegant solution for a long-standing challenge, but in certain cases, it wasn’t quite perfect. The problem, he said, is not blending between mip levels but a filter transition within a mip level. The transition between two kernels doesn’t quite happen as it should. Demers was adamant that Cypress does not cheat on its level-of-detail calculations (a common performance optimization) and that the issue is simply the result of a mistake. Fortunately, the error has been corrected in the Barts filtering hardware, and the result is much smoother transitions.


Nvidia’s texture filtering algorithm strikes a somewhat different balance, as you can see below. (All of these tests were produced using the default filtering quality in the video drivers.)


On the Radeons, the checkerboard pattern melts into a gray circle well before the end of the cylinder, whereas the GF104 shows detail all the way to the end, with some intriguing and intricate moire patterns. Those patterns are smoother and more regular than the ones on the Radeons, which translates into less visual noise. The odd thing about the Nvidia result here is that puffy, uh, donut shape in there. (Mmmmm… donuts.) Heck, the donut isn’t perfectly round, either; it’s more octagonal. Switching on colored mip levels will give us a better sense of what’s happening.




Now the donut on the GeForce looks like a big, red stop sign, which highlights the fact that Nvidia applies a little less filtering to objects at certain angles of inclination. Coloring the mip levels also reveals clearly that Nvidia does less accurate blending between mip levels than AMD, which is what causes the donut effect. The color gradients are much finer on the Radeons, and those smoother transitions produce no visible rings in our high-frequency checkerboard sample.

Which is better overall? I’m not sure I can say, and this is a single, static example that’s very tough to handle. In games, the differences between the GPUs are much less readily evident. The reality is that AMD and Nvidia appear to be very closely matched—even more so now that Barts fixes that filter transition problem.

Morphological AA: sounds cool, right?
The other new image quality enhancement arriving with Barts is a software-based antialiasing filter that AMD has dubbed morphological AA. AMD has been playing around with various custom, post-process AA filters for some time now, and this new one is the next step. Unlike some of AMD’s past custom filters, morphological AA is based on a compute shader and, as I understand it, simply looks at a finished image, detects edges with rough transitions, and attempts to smooth them.

The advantages of this approach are several. The morphological filter has relatively low performance overhead, and because it’s simply looking at a finished scene and doing its work, it will smooth out rough transitions even if they don’t occur along polygon boundaries. The most widely used AA method, multisampling, simply won’t address jagged edges within textures or the like. Also, because it’s a post-process effect, morphological AA should be compatible with a wide range of games based on DX9 and up. Since this feature is implemented in DirectCompute, it will be available via a driver update for owners of existing 5000-series Radeons, as well as the 6850 and 6870.

AMD only got us a driver with morphological AA enabled a couple of (very busy) days ago, so we haven’t had much time to play with it and form many impressions. We have produced some sample images, shown below, from morphological AA versus multisampling. You should know that the morphological AA sample image was produced by a tool from AMD that applies the filter to a screenshot. We had to use this tool because the post-process filter’s effects don’t show up in screen captures.

No antialiasing

4X multisampling

Morphological AA (without MSAA)

8X multisampling

These images from Bad Company 2 are a pretty good example of the problems with multisampled AA. Quite a few of the object edges in this shot aren’t touched by MSAA, regardless of the strength of it. That seems to be a quirk of a lot of modern game engines, including this one. The result is that the sight and the top of the player’s gun are smoothed nicely by MSAA, but very little else in the image is—not the foliage, nor the edge of the cliff curving through the top right portion of the image. By contrast, with morphological AA, a great many of the object edges in the scene are softened.

One disadvantage of morphological AA is that, since the filter operates on a completed scene, it lacks any sort of sub-pixel precision. The edge-detection algorithm must simply guess about the angles of the slopes it’s smoothing. Look, for instance, at the right side of the middle tine of the sight on the player’s gun above. Without AA, that tine is three pixels wide most of the way down and then fans out into a fourth pixel on the right. The morphological filter turns that into a fairly pronounced angled edge, while multisampling (especially at 8X) reveals that line runs very nearly straight up and down.

Ok, if you don’t see that, I don’t blame you. It’s a bit subtle, but the limitation is a real one, and it may mean that object edges tend to crawl or warp while in motion. We need to spend more time with this feature to get a fuller impression of its worth.

The Catalyst drivers for the new Radeons have a couple of other changes, as well. In addition to morphological AA, the older edge-detect CFAA filter remains an option, but the narrow and wide tent filters are not available. Nalasco tells us the decision to remove the tent filters was a consequence of some performance improvements AMD recently made to its edge-detect filter that rendered the tent filters superfluous.

AMD has added another checkbox in the Catalyst Control Center titled “Disable surface format optimization.” This language refers to the fact that AMD’s drivers have been, for certain games, converting HDR textures into lower-precision formats in order to raise performance. Nvidia has been banging the drum about this issue for a while now, saying that it would never do such a thing. Only a handful of games seem to be affected, none of which we’ve used recently for performance testing, so we haven’t spent much time worrying about it. (The list includes Dawn of War II, Empire Total War, Need for Speed: Shift, Oblivion, Serious Sam II, and the original Far Cry.) AMD claims image quality is not visibly reduced by this change, but it has decided to make the concession of letting the user disable this optimization if he wishes. Given the choice, we would prefer to let the game developers choose the appropriate texture formats, so we conducted all of our testing with this checkbox ticked.

Some display and multimedia changes
Although Barts hasn’t changed much in the 3D department, it has seen some revisions in other places, including its display and video playback hardware. AMD has made quite a bit of hay out of being able to support three or more monitors with a single GPU over the past year, particularly in relation to multi-screen gaming with Eyefinity. We’re not suprised, then, to see the firm pushing ahead with support for newer display output capabilities.

The most noteworthy change here is probably support for version 1.2 of the DisplayPort standard. This version has twice the bandwidth of 1.1 and enables some novel capabilities. One of those is transport for high-bitrate audio, including the Dolby TrueHD and DTS Master Audio formats, over DisplayPort. Another is the ability to drive multiple displays off of a single output, either via daisy chaining or the use of a break-out DisplayPort hub with multiple outputs. In one example AMD shared with us, a hub could sport four DVI outputs and drive all of those displays via one DP input. What’s more, Barts-based Radeons can support multiple display timings and resolutions over a single connector, so there’s tremendous flexibility involved. In fact, for this reason, AMD has no need to offer a special six-output Eyefinity edition of the Radeon HD 6870.

The output array on XFX’s Radeon HD 6870

Barts’ standard port array will sport two mini-DisplayPort 1.2 outputs, a single HDMI output, one dual-link DVI port, and one single-link DVI output. Yes, AMD has chosen to drop dual-link support on that second DVI output in favor of more DisplayPort connectivity, and yes, that move seems to be a bit premature to us. A Barts-based card can drive a second dual-link DVI monitor using a DisplayPort to DVI converter, but the dual-link versions of those dongles tend to be relatively pricey. Then again, so are monitors that require dual-link inputs.

Speaking of expensive things, that other port on the card above supports HDMI 1.4a, so it’s compatible with stereoscopic televisions and can be used for the playback of Blu-ray discs in stereoscopic 3D. One other modification of note to the Barts display output hardware is an update to its color gamut correction that should allow for more accurate color representation on wide-gamut panels.

If you do plan to slip on a pair of funny glasses in your living room in order to watch a movie, you should be happy to hear that the Barts UVD block can now decode the MVC codec (used for Blu-ray 3D discs) in hardware, along with the MPEG4 (DivX/Xvid) codec. AMD has also extended its support for MPEG2 acceleration to the entropy decode stage of the video pipeline, further unburdening the CPU. No UVD update would be complete without some improvements to the post-processing routines used to address problems with low-quality source video, and AMD hasn’t left us hanging there, either. What’s more, at its 6800-series press event, AMD had representatives from CyberLink, ArcSoft, and DivX on hand to pledge support for the new UVD hardware in their respective media player programs.

Beyond those changes, the folks in AMD marketing have been working overtime on some crucial marketing name modifications, playing off of the success of “Eyefinity” as a semi-clever play on the word “eye.” We now have EyeSpeed and EyeDef, although I kind of get fuzzy when it comes to tying those things to GPU attributes. I do know that AMD’s Stream Computing initiative has been renamed as AMD Accelerated Parallel Processing Technology, which has a lot more vowels and consonants.

On the initiative front, AMD has decided to counter Nvidia’s 3D Vision push by partnering with third-party makers of shutter glasses, stereoscopic displays, and middleware that adds stereoscopic 3D support to current games. These activities will take place under the “HD3D” banner. We’re a little bit unsure what to make of this effort, for a host of reasons, including the simple fact that we’re dubious on the long-term prospects for glasses-based stereoscopic display schemes. We also have a pretty strong impression that the GPU makers will need to support stereo 3D actively and work directly with game developers in order to get really good results. Middleware vendors like DDD, one of AMD’s new partners, don’t help their case when they claim, for instance, that their TriDef Media Player “automatically” converts 2D source DVD, photos, and videos to stereoscopic 3D. Cardboard cutouts, ahoy! On the flip side, we expect AMD isn’t investing too much in stereoscopic support by cobbling together an initiative like this one. If stereo 3D schemes prove unpopular with consumers, they’ll have less to lose. In other words, Nvidia is grabbing failure—or success—by its bare hands, while AMD is using robotic arms.

Introducing the Radeon HD 6850 and 6870
Now that we’re several thousand words into our review, let’s talk about the new Radeon graphics cards. Here are the most relevant specs:




Radeon HD 6870

900 1120 56 32 4.2 Gbps 256 19W/151W $239
Radeon HD 6850

775 960 48 32 4.0 Gbps 256 19W/127W $179

Both cards have 1GB of GDDR5 memory onboard, and AMD has chosen to de-tune the 6850 only by reducing the number of active SIMD engine/texture unit pairs from 14 to 12 and lowering clock speeds.

Notice the very nice prices in the table above. These cards amount to a major overhaul of the value proposition in the middle of AMD’s lineup.

Peak pixel
fill rate

Peak bilinear
INT8 texel
filtering rate*
*FP16 is half rate


Peak shader

Radeon HD 5830

12.8 44.8 128.0 1792
Radeon HD 6850

24.8 37.2 128.0 1488
Radeon HD 5850

23.2 52.2 128.0 2088
Radeon HD 6870

28.8 50.4 134.4 2016

At $179, the Radeon HD 6850 essentially replaces the weak-sister Radeon HD 5830, yet it has twice the ROP rate of that rather unfortunately crippled Cypress derivative, is based on a smaller board with more modest power consumption, and, well, we’ll show you performance shortly. At $239, the Radeon HD 6870 supplants the Radeon HD 5850, yet the newer card costs less, has a higher ROP rate and slightly more memory bandwidth, with comparable specs otherwise.

Above are a couple of Radeon HD 6870 cards from XFX and Sapphire, both of which are available now, and both of which look to be based on AMD’s reference design. 6870 cards require dual 6-pin auxiliary power connectors, and at 9.75″ long, they should be considerably easier to cram into a case than the over-11″ Radeon HD 5870. They’re practically the same size as the Radeon HD 5850, though.

Pictured above is the reference version of the Radeon HD 6850, which has a single 6-pin power input and measures 9″ long.

Right out of the gate, XFX is offering a verison of the 6850 based on its own custom board design that’s a quarter-inch shorter than AMD’s and rocks a Zalman-esque dual-heatpipe cooler. XFX says it’s using higher-quality components to give its cards longer life and better overclocking headroom.

Unfortunately, both of these XFX cards are currently selling for 20 bucks above AMD’s suggested price at Newegg, while other versions, like this Sapphire 6850, are priced in line with AMD’s guidance. XFX may be able to command something of a premium thanks to its lifetime warranty and solid reputation for support, but $20 seems like a lot to ask.

From left to right: Radeon HD 5870, 6870, and XFX’s custom 6850

The competition sharpens its swords
Of course, Nvidia wasn’t about to let AMD introduce new graphics cards without a welcoming committee. The greeting ceremony for Barts arguably started several weeks ago, when Nvidia dropped the price of GeForce GTX 460 768MB cards to $169. Then things got weird, as AMD held out on revealing Radeon HD 6800-series pricing to reviewers until earlier this week. Nvidia then slashed its prices quicker than a fireworks tent on the fifth of July, taking the GeForce GTX 460 1GB down to $199 and dropping the GTX 470 to $259.

That maneuver prompted an amusing back-and-forth in which AMD sent out a retaliatory e-mail claiming Nvidia’s price cuts were only temporary—complete with a promotional directive from Nvidia written in French as ostensible proof. Nvidia responded by saying, essentially, “Nuh uh!” and insisting its price cuts are permanent. We’ll take them at their word for now, but hold them to it later.

In fact, Nvidia tells us to expect the cards available at $199 to be somewhat faster than the the GTX 460’s original 675MHz base clock, as its partners de-emphasize the lower-clocked models over time. The MSI Cyclone card pictured above is right at $199.99 at Newegg at present, and we’ve included in our tests over the following pages.

Not only that, but Nvidia and its board partners have equipped us with a handful of intriguing new GeForces in the past week.

This imposing fellow is a GeForce GTX 460 1GB card from MSI with an 810MHz core clock and 3.9 Gbps memory. That’s positively stratospheric compared to the initial 675MHz core and 3.6 Gbps memory of the GTX 460, and the higher frequencies should translate pretty directly into stronger performance. Better still, this version of the card, known rather comically as the Hawk Talon Attack, features 0.4-ns GDDR5 memory that promises additional overclocking headroom. We haven’t yet had time to test its limits, but between the RAM, the dual-fan cooler, and the fact that MSI’s software offers overvolting of the GPU core and memory, yeah, we’d like to try soon. This puppy is going for $215 at Newegg right now, not far above the GTX 460 1GB’s base price. We have a full set of performance results for this card.

If you can’t be bothered to overclock a graphics card yourself and 810MHz just isn’t enough, there’s this unassuming little number from EVGA, the GeForce GTX 460 1GB FTW edition. Currently selling for $229, this GTX 460 1GB is clocked at a nosebleed-inducing 850MHz with 4 Gbps memory.

There is precedent for this sort of clock speed and performance creep in Nvidia graphics card models over time. Heck, the GeForce GTX 260 actually transitioned from 192 to 216 ALUs and saw prevailing speeds rise from a 576MHz base to 650MHz and more during its run. Still, this is quite the promotion for the humble GTX 460 in a pretty short span.

Time limits prevented us from testing the EVGA FTW edition in our full suite, but we only left it out of a couple of the games we tested manually with FRAPS.

The final member of the Barts welcoming committee is this GeForce GTX 470 from Galaxy. This “GC Edition” card boasts a modest clock speed bump from Nvidia’s stock 607MHz to 625MHz, but its real appeal is a blue PCB and that slick looking plastic cooling shroud that looks like it ought to have little green army men hiding inside of it. In fact…

There’s a hatch for the army men to hide under! Sweeeeet.

Actually, the product packaging makes no mention of why the fan flips up, but I’ve heard it’s purported to be for easy cleaning of dust and lint. Whatever the case, this cooler is certainly distinctive. This beast isn’t currently listed at Newegg, but Asus and others have stock-clocked GTX 470s at Nvidia’s new $259 suggested price. EVGA also has a 625MHz variant for $269.

Our testing methods
Many of our performance tests are scripted and repeatable, but for some of the games, including Battlefield: Bad Company 2 and Mafia II, we used the Fraps utility to record frame rates while playing a 60-second sequence from the game. Although capturing frame rates while playing isn’t precisely repeatable, we tried to make each run as similar as possible to all of the others. We raised our sample size, testing each FRAPS sequence five times per video card, in order to counteract any variability. We’ve included second-by-second frame rate results from Fraps for those games, and in that case, you’re seeing the results from a single, representative pass through the test sequence.

As ever, we did our best to deliver clean benchmark numbers. Tests were run at least three times, and we’ve reported the median result.

Our test systems were configured like so:

Processor Core i7-965 Extreme 3.2GHz
Motherboard Gigabyte EX58-UD5
North bridge X58 IOH
South bridge ICH10R
Memory size 12GB (6 DIMMs)
Memory type Corsair Dominator CMD12GX3M6A1600C8
at 1600MHz
Memory timings 8-8-8-24 2T
Chipset drivers INF update
Rapid Storage Technology
Audio Integrated ICH10R/ALC889A
with Realtek R2.51 drivers
Graphics Gigabyte
Radeon HD 4850 OC 1GB
with Catalyst 8.782-100930m drivers & 10.9a application profiles
Radeon HD
4870 1GB
with Catalyst 8.782-100930m drivers & 10.9a application profiles
XFX Radeon HD
5830 1GB
with Catalyst 8.782-100930m drivers & 10.9a application profiles
Radeon HD 5850 1GB
with Catalyst 8.782-100930m drivers & 10.9a application profiles
Asus Radeon HD
5870 1GB
with Catalyst 8.782-100930m drivers & 10.9a application profiles
XFX Radeon HD
6850 1GB
with Catalyst 8.782-100930m drivers & 10.9a application profiles
XFX Radeon HD
6870 1GB
with Catalyst 8.782-100930m drivers & 10.9a application profiles
Asus GeForce
GTX 260 TOP SP216 1GB
with ForceWare 260.89 drivers
GeForce GTX 460 768MB OC
with ForceWare 260.89 drivers
MSI Cyclone
GeForce GTX 460 1GB 725MHz
with ForceWare 260.89 drivers
MSI Hawk
Talon Attack GeForce GTX 460 1GB 810MHz
with ForceWare 260.89 drivers
EVGA GeForce GTX 460 1GB FTW 750MHz
with ForceWare 260.89 drivers
Galaxy GeForce GTX 470 1280MB GC
with ForceWare 260.89 drivers
GeForce GTX 480 1536MB
with ForceWare 260.89 drivers
Hard drive WD RE3 WD1002FBYS 1TB SATA
Power supply PC Power & Cooling Silencer 750 Watt
OS Windows 7 Ultimate x64 Edition
DirectX runtime update June 2010

Thanks to Intel, Corsair, Western Digital, Gigabyte, and PC Power & Cooling for helping to outfit our test rigs with some of the finest hardware available. AMD, Nvidia, and the makers of the various products supplied the graphics cards for testing, as well.

Unless otherwise specified, image quality settings for the graphics cards were left at the control panel defaults. Vertical refresh sync (vsync) was disabled for all tests.

We used the following test applications:

The tests and methods we employ are generally publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.

Running the numbers

Peak pixel
fill rate

Peak bilinear
INT8 texel
filtering rate*
*FP16 is half rate


Peak shader

GeForce GTX 260 TOP SP216

18.2 46.8 128.8 605 650
GeForce GTX 460 768MB

16.2 37.8 86.4 907 1350
GeForce GTX 460 1GB

21.6 37.8 115.2 907 1350
GeForce GTX 460 1GB 725MHz

23.2 40.6 115.2 974 1450
GeForce GTX 460 1GB 810MHz

25.9 47.6 124.8 1089 1620
GeForce GTX 465

19.4 26.7 102.6 855 1821
GeForce GTX 470

24.3 34.0 133.9 1089 2428
GeForce GTX 470 GC

25.0 35.0 133.9 1120 2500
GeForce GTX 480

33.6 42.0 177.4 1345 2800
Radeon HD 4850 OC

11.2 28.0 63.6 1120 700
Radeon HD 4870

12.0 30.0 115.2 1200 750
Radeon HD 5770

13.6 34.0 76.8 1360 850
Radeon HD 5830

12.8 44.8 128.0 1792 800
Radeon HD 5850

23.2 52.2 128.0 2088 725
Radeon HD 5870

27.2 68.0 153.6 2720 850
Radeon HD 6850

24.8 37.2 128.0 1488 775
Radeon HD 6870

28.8 50.4 134.4 2016 900
Radeon HD 5970

46.4 116.0 256.0 4640 1450

We’ve already looked at how the 6850 and 6870 compare to the 5850 and 5870, but here’s a broader comparison of specs. As always, these are just theoretical peaks and don’t necessarily predict delivered performance.

Nvidia and AMD do largely seem to be converging on a common target in terms of resource balance. Take the 6850 and the GTX 460 725MHz, for instance. They have very similar peak ROP/pixel fill and texture filtering rates, although the 6850 has a little more memory bandwidth. As usual, the Radeon has a substantially higher theoretical peak arithmetic rate, but Nvidia’s GPUs seem to be efficient enough at executing actual shader code to overcome that gap. Also, relatively speaking, that shader rate gap is shrinking. The 5830’s peak is about 1.8 teraflops, but the 6850 peaks at just under 1.5 teraflops.

We like to see how well these GPUs can approach their theoretical best peformance with directed tests when possible. We’ve already looked at tessellation performance earlier, so here’s a quick look at texture sampling and filtering capabilities.

We’ve grown increasingly dissatisfied with the texture fill rate tool in 3DMark Vantage, so we’ve reached back into the cupboard and pulled out an old favorite, D3D RightMark, to test texture filtering performance. Unlike 3DMark, this tool lets us test a range of filtering types, not just texture sampling rates. Unfortunately, D3D RightMark won’t test FP16 texture formats, but integer texture formats are still pretty widely used in games. I’ve plotted a range of results below, and to make things more readable, I’ve broken out a couple of filtering types into bar charts, as well.

In theory, the newer Radeons have slightly lower texturing capacity than the products they replace, and that holds up in our measurements, as the 6850 just trails the 5830 and the 6870 does the same versus the 5850. By contrast, the full-on Cypress aboard the Radeon HD 5870 is a titan of texture filtering prowess, even faster than the mighty GTX 480. Still, the cards based on the trimmed-back Barts GPU retain rates that are quite competitive with those of their GeForce counterparts.

Battlefield: Bad Company 2
BC2 uses DirectX 11, but according to this interview, DX11 is mainly used to speed up soft shadow filtering. The DirectX 10 rendering path produces the same images.

We turned up nearly all of the image quality settings in the game. Our test sessions took place in the first 60 seconds of the “Heart of Darkness” level.

Our very first game benchmark gives us a feel for the added efficiency of the Barts GPU. Despite the trimming to Barts’ shader and texturing power, the Radeon HD 6870 not only keeps up with the 5850, but matches the 5870 frame for frame. The 6850, meanwhile, relegates the 5830 to its proper status as a bad memory. This thing is a much more competitive product.

Speaking of competition, the new Radeons acquit themselves nicely versus those pesky GeForces. The 6850 slots in right between the GTX 460 768MB and the 1GB 725MHz version, as does its pricing, and the 6870 essentially ties with the very fastest 850MHz variant of the GTX 460 1GB—and with the pricier GTX 470.

Owners of older Radeon HD 4800-series cards will want to consider these fresh Radeons carefully. The 6850 offers a 50% increase in measured frame rates over ye olde 4850, and more importantly, it takes this game from sluggish to smooth at the very common 1080p resolution.

Starcraft II
Up next is a little game you may have heard of called Starcraft II. We tested SC2 by playing back a match from a recent tournament using the game’s replay feature. This particular match was about 10 minutes in duration, and we captured frame rates over that time using the Fraps utility. Thanks to the relatively long time window involved, we decided not to repeat this test multiple times, like we usually do when testing games with Fraps.

We tested at the settings shown above, with the notable exception that we also enabled 4X antialiasing via these cards’ respective driver control panels. SC2 doesn’t support AA natively, but we think this class of card can produce playable frame rates with AA enabled—and the game looks better that way.

The new Radeons perform admirably once more, ever-so slightly outgunning the closest competition from the green team. Most of these cards deliver eminently acceptable frame rates at this extreme display resolution, but the two slowest cards obviously stumble—the 5830 likely due to its weak ROP rate, while the GTX 460 768MB is probably bumping up against a video memory limitation.

Aliens vs. Predator
The new AvP game uses several DirectX 11 features to improve image quality and performance, including tessellation, advanced shadow sampling, and DX11-enhanced multisampled anti-aliasing. Naturally, we were pleased when the game’s developers put together an easily scriptable benchmark tool. This benchmark cycles through a range of scenes in the game, including one spot where a horde of tessellated aliens comes crawling down the floor, ceiling, and walls of a corridor.

To keep frame rates playable on these cards, we had to compromise on image quality a little bit, mainly by dropping antialiasing. We also held texture quality at “High” and stuck to 4X anisotropic filtering. We did leave most of the DX11 options enabled, including “High” shadow quality with advanced shadow sampling, ambient occlusion, and tessellation. The use of DX11 effects ruled out the use of older, DX10-class video cards, so we’ve excluded them here.

Another page of results reinforces our sense that the 6800-series Radeons have hit their marks. They also confirm that the competing GeForce cards are right in the mix. This is not going to be an easy one to call, is it?

Metro 2033
The developers of Metro 2033 have come up with a nifty scripted benchmark based on their game, and we decided to give it a shot. As the settings page below shows, we did not test at Metro 2033‘s highest image quality settings. Those wouldn’t be too friendly to mid-range graphics cards like these mostly are. Because we used DirectX 11, we had to exclude the older cards from this one.

Wow, these results are a little top-heavy with green. That’s true in part because we’ve included the single biggest DX11 GPU and the most expensive single-GPU graphics card you can buy, the $500 GeForce GTX 480. Even so, the GeForces tend to be quite a bit faster in this game—with the obvious exception of the GTX 460 768MB at 2560×1600. That lower-memory card looks to be a poor match for a high-resolution display.

Why are the Barts cards faster than the Cypress ones at 2560×1600? I dunno, but we’ll be entertaining guesses in the comments. Perhaps it has to do with tessellation overhead, which would also explain why the GF100-based cards, the GTX 470 and 480, so handily outrun their GF104-based brethren.

DiRT 2: DX9
This excellent racer packs a scriptable performance test. We tested at DiRT 2‘s “ultra” quality presets in both DirectX 9 and Direct X 11. The big difference between the two is that the DX11 mode includes tessellation on the crowd and water. Otherwise, they’re hardly distinguishable.

DiRT 2: DX11

The GeForces handle this game particularly well at lower resolutions and in DirectX 9. As the display resolution rises and we layer on DX11 features, though, the balance shifts back toward the Radeons. In the end, we’re back to near performance parity at the most demanding settings.

Like the narrator in a National Geographic special, we should pause with faux empathy to note a rather unfortunate act of cannibalization: the higher-clocked GTX 460 cards outperform the GTX 470 here. This same thing nearly happened in our Bad Company 2 tests, but now the old girl has finally succumbed. Thus the circle of life is complete. Or, wait, maybe I’m supposed to blame global warming? I forget how this goes.

We tested Gearbox’s post-apocalyptic role-playing shooter by using the game’s built-in performance test with all of the in-game quality options at their max. We didn’t enable antialiasing, because the game’s Unreal Engine doesn’t natively support it.

We’ve included Borderlands once again because it’s one of my favorite games ever, the test is easily scriptable, the game itself tends to be pretty demanding on fast video cards, and it’s based on the oh-so-popular Unreal engine. However, we’ve watched GeForces stomp on Radeons in this test so many times now, it may be time to move on. I just wish AMD would give this game a little bit of tuning attention in its drivers somehow. Heck, the ol’ GTX 260 is nearly as fast here as the Cypress cards.

One thing we can say is that AMD’s architectural compromises with Barts appear to have paid off for this game. The Barts-based 6870 matches the Cypress-fortified 5870 almost exactly.

Mafia II
The open-world Mafia II is another new addition to our test suite, and we also tested it with Fraps.

If you turn on antialiasing in this game, that apparently does something unexpected: enables a 2X supersampled antialiasing mode. Supersampling touches every single pixel on the screen and thus isn’t very efficient, but we still saw playable enough frame rates at the settings we used. In fact, we need to look into it further, but we think Mafia II may also be using some form of post-processing or custom AA filter to further soften up edges. Whatever it’s doing, though, it seems to work. The game looks pretty darned good to our eyes, with very little in the way of crawling or jaggies on edges.

Although this game includes special, GeForce-only PhysX-enhanced additional smithereens and flying objects, we decided to stick to a direct, head-to-head comparison, so we left those effects disabled.

You’ll notice that most of the GeForces are missing from the results above. That’s because we found a pretty serious problem with how we’d tested this game once we’d compiled the results. Have a look at the frame rate lines for the faster GeForce cards below:

For most of our gaming session, the GeForces look to be effectively capped at a 60 FPS frame rate. That’s not entirely the case, as the section of the end of each run—a quick scripted sequence in the game engine—demonstrates; the GeForces range well over 60 FPS then. For whatever reason, though, they’re stopping at 60 FPS otherwise. As a result, we’ve excluded the faster GeForce cards from the main results above, and we’re not entirely confident about the numbers for the slower cards, either. We’ve just left in these Mafia II results so you can compare the Radeons to one another.

On that front, the 6850 and 6870 perform admirably, with the 6870 once again topping even the 5870.

Power consumption
We measured total system power consumption at the wall socket using our fancy new Yokogawa WT210 digital power meter. The monitor was plugged into a separate outlet, so its power draw was not part of our measurement. The cards were plugged into a motherboard on an open test bench.

The idle measurements were taken at the Windows desktop with the Aero theme enabled. The cards were tested under load running Left 4 Dead 2 at a 1920×1080 resolution with 4X AA and 16X anisotropic filtering. We test power with Left 4 Dead 2 because we’ve found that the Source engine’s fairly simple shaders tend to cause GPUs to draw quite a bit of power, so we think it’s a solidly representative peak gaming workload.

Oh, and our graph labels have changed on this page to more fully reflect the brands of the cards being used, since custom board designs and coolers will have a major impact on power, heat, and noise. We tested only the XFX version of the 6870 here because it, the Sapphire, and the reference card from AMD all share the same cooler and board layout.

AMD led us to expect some nice reductions in idle power use with the Barts-based cards, but we just didn’t see much on our power meter. Nvidia’s GTX 460 cards still draw appreciably less power when they’re not busy.

Barts does draw less power under load than Cypress, and Barts is also more power-efficient than the GTF104 at comparable performance levels. Those GTX 460s clocked at over 800MHz cause our test rig to pull 12W more than it does with a 6870 in the PCIe slot.

Noise levels
We measured noise levels on our test system, sitting on an open test bench, using an Extech model 407738 digital sound level meter. The meter was mounted on a tripod approximately 10″ from the test system at a height even with the top of the video card.

You can think of these noise level measurements much like our system power consumption tests, because the entire systems’ noise levels were measured. Of course, noise levels will vary greatly in the real world along with the acoustic properties of the PC enclosure used, whether the enclosure provides adequate cooling to avoid a card’s highest fan speeds, placement of the enclosure in the room, and a whole range of other variables. These results should give a reasonably good picture of comparative fan noise, though.

The differences in noise levels at idle for most of these cards aren’t sufficiently large for us to say with any confidence that you’d notice them. We’re simply sticking a sound level meter on a tripod next to a test system and recording a number; we don’t have Steve Jobs’ massive ana-whatever chamber buried six miles deep to ensure total isolation. Only the GTX 480 and better are likely to be appreciably noisier, with the one real stand-out here (in a bad way) being XFX’s take on the Radeon HD 6850.

Wow, things are all over the map here, and noise levels don’t seem to track with peak power consumption levels as one might expect. The Radeon HD 6850’s reference cooler has hit the same basic target as the two 5800-series cards did before it, and that’s not too bad. There are much quieter GeForces, though, including the Galaxy GTX 470 and EVGA’s 850MHz GTX 460 FTW—both of which are very impressively quiet for their power draw levels.

Several cards are unfortunately on the noisy side, including the MSI Hawk Talon Attack, XFX’s 6850, and the Radeon HD 6870’s common reference cooler. I have a good sense of what’s happening with the other two cards, as we’ll discuss below, but the 6870’s noise levels are a little disappointing for a card that draws so much less power than a Radeon HD 5870—or, jeez, a GTX 480.

GPU temperatures
We used GPU-Z to log temperatures during our load testing.

Notice the presence of the two MSI cards and the XFX Radeon HD 6850 at the top of the chart above, indicating they have the lowest operating temperatures. Board makers appear to be tuning their custom coolers these days to achieve especially low temperatures, even if it means they’ll be several decibels louder than necessary. When we’ve asked about these tuning decisions in the past, the motivations seem to be related to creating more overclocking headroom or extending the life of the GPU silicon. We can’t say we like it, though, when a snazzy and obviously effective cooler like the quad-heatpipe number on the MSI Talon Attack registers two decibels above a GTX 480 on our meter. We can’t help but get the sense that board makers don’t share our priorities when they build in such noisy default fan speed profiles. Meanwhile, the stock cooler on EVGA’s 850MHz GTX 460 is whisper quiet and keeps temperatures well within a reasonable range.

Fortunately, the folks at XFX tell me they are considering offering their users a choice by making an alternative BIOS with a quieter fan profile available on their website for this 6850. If that happens, we’ll try to grab it and see how it handles.

These two new Radeons are surgical strikes at the competition, not all-out assaults intended to alter the balance of power massively. AMD has simply taken a proven technology, refined it slightly, and targeted it at a couple of crucial price points in the middle of the market. Even the few tweaks AMD’s engineers made to the GPU’s 3D graphics hardware—the filtering fix and the tessellation optimizations—mainly just help even things up with Nvidia’s GF104.

Happily, though, both of these Radeons have hit their marks, bringing enough performance and capability to their respective targets to make not just the parts they replace but all of the Radeon HD 5800-series cards feel a bit pointless. Yes, the 5870 is still a bit faster than the 6870, but not by much. The balance of graphics hardware resources in Barts looks to be golden for today’s games; the cuts to shader and texturing power barely sting.

The Radeon HD 6850 and 6870 are also good enough to have forced Nvidia’s hand on a frantic round of clock speed increases and price cuts, and we’re naturally pleased to see it happen. Nvidia’s adjustments do currently seem to be sufficient to keep the various versions of the GTX 460 we tested competitive, price-performance wise, with the 6850 and 6870. I’m not sure I could choose between the two GPU brands right now in this category, it’s so even. If there is a mismatch here, it’s between the GTX 460 768MB at $169 and the 6850 at $179. That 768MB card seems to run out of video memory at higher display resolutions, so unless you’re absolutely married to a single, lower-resolution monitor, we would recommend paying the extra 10 bucks for the 6850.

Then again, we’re bickering over $10 here, and I’ve sworn that off in the past. Video card prices do tend to drop around the introduction of a new product, and the landscape may change dramatically in the coming days and weeks. What you should do if you want a new video card for gaming is use the 6850 and the GTX 460 1GB as your new baselines. Don’t bother with cheaper cards like the Radeon HD 5770 or the GeForce GTS 450 unless you’re truly strapped for cash. The deals are too good to ignore at around $180.

The only thing that might keep you from snapping up one of these cards now is, well, even higher aspirations. If you’re hoping for a baby Barts to push the envelope at under $150, don’t hold your breath. AMD tells us the Juniper GPU in the Radeon HD 5700 series will be sticking around for a while. If you’re looking for a true and proper replacement for the Radeon HD 5870 that ups the ante on performance, however, you may want to hold out for a bit. Barts is just part of AMD’s multi-pronged approach to refreshing its GPU lineup, and a larger, more potent, and rather intriguing chip code-named Cayman is just over the horizon. We’ve also heard whispers of a dual-Cayman product code-named Antilles that could be stupendously quick. These Northern Islands are full of surprises.

Scott Wasson

Latest News

Twitter Reveals a Part of Its Code for Public Inspection

Twitter Reveals a Part of Its Code for Public Inspection

“AI Developments Should Be Paused for Half-a Year”: AI Ethicists

“AI Developments Should Be Paused for Six Months”: AI Ethicists

On March 28th, 2023, a group of acclaimed ethicists published a letter from the Future of Life Institute, pleading for a six-month moratorium on AI development projects. Signed by more...

T2 Challenges Twitter With Its New Verification Program

T2 Challenges Twitter With Its New Verification Program

The seed-funded Twitter rival, T2, introduces a new verification program. While Elon’s Twitter has focused on making money through its services, T2 is seemingly trying to use the loophole. The...

AI Research Group Urges FTC to Stop OpenAI’s GPT Releases

AI Research Group Urges FTC to Stop OpenAI’s GPT Releases

deelance 5
Blog, Crypto News, News

DeeLance Price Prediction – This Web 3.0 Decentralized Freelancer Project’s Presale Will Explode!

Blog, Crypto News

The 6 Best Cryptos To Buy On Presale This Week!

ecoterra presale
Blog, Crypto News, News

5 reasons Brand New Green Crypto, Ecoterra Will Explode on Presale!