by Sergey Lepilov
11/16/2010 | 11:26 PM
According to statistical data collected by the Steam system, multi-GPU configurations are very far from popular. They are only present in computers of less than 1 percent of all gamers. Interestingly, 96% of such configurations are based on Nvidia’s SLI technology whereas AMD’s CrossFireX only amounts to 4%. Of course, Steam collects data for the limited number of games and applications that it distributes or supports. Even though Steam delivers many bestselling titles, its statistics cannot be viewed as the final truth. On the other hand, there are no other statistical data available on that issue.
It doesn’t take much insight into the market situation to realize that multi-GPU configurations are indeed very, very scarce. There are very few users of two graphics cards on hardware forums and, try as you might, you can’t virtually find any people who build graphics configurations with three or four cards. Besides the obvious reason of high price (including the higher price of compatible mainboards and high-wattage power supplies), the deterring factors are the low stability and efficiency of SLI and CrossFireX subsystems which depend greatly on software optimizations.
However, we have noticed a rather optimistic trend during this year as multi-GPU technologies seem to have improved. This is especially true for CrossFireX which used to have much more stability and efficiency issues than SLI (by the way, the latter technology offers the unique opportunity to choose from different rendering modes in the driver and thus can work in more games). AMD has been really working hard to improve CrossFireX. For example, they have recently introduced CrossFireX profiles to install together with the Catalyst driver and be updated regularly. And like you will see in our today's tests, there are also hardware changes in the new Radeon HD 6850 and 6870 cards.
So, this review is all about the compatibility and efficiency of CrossFireX technology as it is now.
The graphics cards in the following table as well as on the results diagrams later on are sorted accoridng to their recommended price (from high to low):
We published a special report about the new graphics cards titled ATI Radeon HD 6800: Generation Next. So in this section we will just give you some more info on the reference Radeon HD 6850 and Radeon HD 6870 models (on the left and on the right, respectively, in this and other photographs):
So, the Radeon HD 6850 is 229 millimeters long and its cooling system doesn’t exceed the dimensions of the PCB. The card’s height is 36 millimeters. The GPU is placed closer to the mounting bracket, the distance from the closest mounting hole of the GPU heatsink to that bracket being 74 millimeters, which makes it impossible to install many alternative coolers.
The Radeon HD 6870 is 242 millimeters long, i.e. 13 millimeters longer than its cousin. Its cooler protrudes beyond the PCB by 7 millimeters. The height is the same 36 millimeters. The GPU is placed closer to the mounting bracket, too, but the distance from its heatsink to the bracket is 87 millimeters (more than on the HD 6850). The diagonal distance between the GPU heatsink's mounting holes is 75 millimeters on both cards, just like on the Radeon HD 5870 and 5850.
The graphics dies of the Radeon HD 6870 and 6850 are the same size of 255 sq. millimeters. They only differ with their markings and manufacturing dates: weeks 37 and 36 of 2010 for our 6850 and 6870 samples, respectively.
We can remind you that the Radeon HD 6850's GPU has 160 unified shader processors and 8 texture-mapping units less than the HD 6870. Its frequency is 775 MHz as opposed to the HD 6870's 900 MHz. Both cards drop their GPU frequency to 100 MHz in power-saving mode. The GPU voltage is also lowered then: from 1.094 V to 0.95 V on the HD 6850 and from 1.172 V to 0.95 V on the HD 6870.
Each card is equipped with 1024 megabytes of GDDR5 memory from Hynix Semiconductor Inc. These H5GQ1H24AFR T2C chips have a rated voltage of 1.5 volts and a rated frequency of 5000 MHz.
All of the chips are placed on the face side of the PCBs. The memory bus is 256 bits wide. The Radeon HD 6850 has a memory frequency of 4000 MHz and the Radeon HD 6870, 4200 MHz.
Here are the parameters of the reference cards:
These cards have completely different coolers:
The GPUs are cooled with small aluminum heatsinks with copper soles. The difference is that the Radeon HD 6850 has an evaporation chamber in the base of the heatsink while the Radeon HD 6870 has a copper plate with grooves for three heat pipes. Two of those pipes are 6 millimeters in diameter and the central one is 8 millimeters.
The blowers of these cards differ in shape (the Radeon HD 6870 has a wider blower) as well as in electrical parameters. The Radeon HD 6850 has an 8.4W FirstD fan (marked as FD9238M12D). The Radeon HD 6870 is equipped with an FD9238Р12S fan from NTK Ltd.
The speed of the fans is regulated automatically through pulse-width modulation. By the way, the former card’s fan is fastened in the cooler’s plastic casing while the latter’s one is secured on a metallic frame. This difference couldn’t but affect the noise level which we will discuss later on. Right now, let’s check out how hot these reference graphics cards are.
Here is how we tested the coolers of these and other cards. We install the graphics card into a closed system case whose configuration is listed in the next part of the review. The ambient temperature is constant at 24°C. The system runs FurMark version 1.8.2 launched from a renamed EXE file at 2560x1600 with 16x anisotropic filtering forced on in the driver. We use GPU-Z 0.4.7 and MSI Afterburner 2.0.1 beta 1/2/3 for monitoring. The test is performed before we take off the graphics card’s cooler and with its default GPU thermal interface.
Here are the results of the two first cards in the automatic fan control mode…
|
|
|
...and at the highest fan speed.
|
|
|
So, the graphics core of the Radeon HD 6850 is as hot as 93°C in the automatic mode while its power components are 74°C hot. The fan speed grew from 1000 to 2435 RPM at that. Despite its higher frequency, the GPU of the Radeon HD 6870 is no hotter than 87°C and the power components not hotter than 82°C at 2430 RPM. The difference in performance between the reference coolers of the Radeon HD 6850 and 6870 is obvious at the maximum fan speed where the Barts Pro is as hot as 78°C and higher even at 4000 RPM whereas the Barts XT is only 62°C hot at 4420 RPM. Well, as a matter of fact, each blower is uncomfortably noisy even at 1900 RPM and downright loud at 2400 RPM and higher.
When in CrossFireX mode, both pairs of graphics cards have even higher temperatures and fan speeds.
|
|
|
Considering the revised PCB layout (the GPUs are now closer to the video interfaces), installing alternative coolers on these graphics cards is problematic. For example, the new Thermalright Shaman can be installed neither on the reference Radeon HD 6850 nor on the reference Radeon HD 6870. The high-performance cooler Accelero XTREME 5870 from Arctic Cooling doesn’t fit on the Radeon HD 6850 but suits the Radeon HD 6870 nicely.
But you can see that some 50 millimeters of the cooler’s length are redundant. So, you can limit yourself to the Arctic Cooling Twin Turbo Pro, especially as it has already acquired official support for the new Radeon series products. When used with an Accelero XTREME 5870, the Radeon HD 6870 became very cool even without heatsinks on its power components and at a quiet speed (1100 RPM) of the three 92mm fans.
AMD Radeon HD 6870
(Accelero XTREME 5870, 3 x 1100 RPM)
AMD Radeon HD 6870
(Accelero XTREME 5870, 3 x 1970 RPM)
At the maximum speed of the three fans, the GPU temperature is only as high as 57°C.
When we tried to overclock our cards, the two samples of Radeon HD 6850 achieved similar GPU frequencies (890 and 910 MHz) without losing their stability or image quality. We didn’t try to overclock them with a voltage increase because we couldn’t provide good enough cooling for them. As for the graphics memory, one sample stopped at a modest 4520 MHz while the other achieved a peak memory frequency of 4880 MHz. The overclocking results of the better sample are shown in the next screenshot:
The two Radeon HD 6870 samples with reference coolers and with default GPU voltages overclocked differently: one sample achieved a higher GPU frequency and the other sample, a higher memory frequency.
Besides the new cards from AMD, this test session will also include two off-the-shelf graphics cards from Palit: GeForce GTX 460 Sonic Platinum 1GB. They come in small but brightly decorated boxes.
You can find exhaustive information about the product on the face and back sides of the box. Inside, there is a central compartment with the graphics card, the accessories lying nearby.
The accessories are very scanty: a power cable, a CD with drivers and a brief installation guide. We wonder why Palit didn’t include something like a SLI bridge, adapters or games with this product. Saving on accessories can make sense, but we don’t think the Palit GeForce GTX 460 Sonic Platinum is a proper case for that.
After we’ve taken a look at the Radeon HD 6850 and HD 6870, the Palit GeForce GTX 460 Sonic Platinum seems very small with its length of only 188 millimeters. Even though it has a dual-slot cooler design, it looks like a toy compared to the new Barts-based products from AMD.
It is quite functional, however, being equipped with one D-Sub, one DVI-I, one DVI-D and one HDMI output.
So, there is everything you need except for DisplayPort. The rest of the card’s mounting bracket is occupied by two vent grids for exhausting the hot air out of the system case.
The graphics card’s cooler casing with a fan impeller is fastened to the PCB separately from the heatsinks on the GPU and power components.
As you can see, the GPU heatsink has a copper base contacting with the GPU heat-spreader via a thick and rather nonuniform layer of thermal grease. It also has two copper heat pipes with a diameter of 6 millimeters and thin aluminum fins. The other heatsink is a simple aluminum thing painted black and equipped with a thermal pad.
The Palit GeForce GTX 460 Sonic Platinum features a custom PCB design and a 4-phase power subsystem.
The two 6-pin power connectors are oriented upwards, so the card proves to be indeed very short and easy to use. However, we want you to note how close the GPU is to the card’s mounting bracket. This may make the card incompatible with some alternative coolers.
Manufactured on Taiwan, the 40nm GF104 chip is covered with a heat-spreader that carries the marking and manufacturing date, which is the 23rd week of 2010.
The GPU works at frequencies of 800/1600 MHz, which is 18.5% above those of the reference GeForce GTX 460. That's quite aggressive factory overclocking, we should say. The rest of the GPU characteristics are the same as those of the reference card.
The Palit GeForce GTX Sonic Platinum has 1 gigabyte of GDDR5 memory on the face side of its PCB. These K4G10325FE-HC05 chips were manufactured by Samsung Semiconductor.
The chips have a fetch time of 5 nanoseconds and a rated frequency of 4000 MHz. This is exactly the frequency the Palit GeForce GTX 460 Sonic Platinum clocks its memory at (11% above the reference card's memory clock rate: the reference GeForce GTX 460 has a memory frequency of 3600 MHz). The memory bus is 256 bits wide.
Thus, these cards from Palit are among the fastest off-the-shelf versions of the GeForce GTX 460 available today.
The Gigabyte GeForce GTX 460 GV-N460SO-1GI is the only version we know of that comes with higher frequencies, 815/4000 MHz. There are also GeForce GTX 460 cards from Zotac and Leadtek that have the same clock rates as the Palit GeForce GTX 460 Sonic Platinum.
The graphics card's heatsink is cooled by a single 92mm fan from Power Logic.
The speed is adjusted automatically through pulse-width modulation within a range of 1200 to 3900 RPM. Let’s see how the cooling system copes with the pre-overclocked card:
Auto mode
Maximum fan speed
The GPU temperature is as high as 83°C in the automatic mode and 80°C at the maximum fan speed. Considering the factory overclocking, this seems to be just fine, but take note of the speed: the fan rotates at 3480 RPM, which is very loud, even in the automatic mode, let alone the maximum speed. Granted, no real game or application can load a graphics card is heavily as FurMark, but the fan speed was about as high as 2900 RPM in games, which can hardly be quiet, either. Running a little ahead, we must confess that the high noise level is the single downside we could find about the Palit GeForce GTX 460 Sonic Platinum.
As for overclocking, both cards, unlike the above-discussed products from AMD, sped up to nearly the same frequencies: 830(840)/1660 MHz for the GPU and 4220 MHz for the graphics memory chips.
This is not much compared to the default frequencies of these cards but quite a lot compared to the frequencies of the reference GeForce GTX 460.
Now let’s check out how hot these cards from Palit are when working in a SLI tandem.
Sadly enough, CrossFireX and SLI configurations call for liquid cooling systems because it is impossible to cool two closely located graphics cards effectively with air coolers at an acceptable level of noise.
We would like to add that Palit GeForce GTX 460 Sonic Platinum is selling for less than $220.
All graphics cards were benchmarked in a closed system case with the following configuration:
I overclocked our 32 nm six-core CPU with the multiplier set at 24x and “Load-Line Calibration” enabled to 4.512GHz. The processor Vcore was increased to 1.475V in the mainboard BIOS:
The 6 GB of system DDR3 memory worked at 1.5 GHz frequency with 7-7-7-14_1T timings and 1.64V voltage. Turbo Boost and Hyper-Threading technologies were disabled during our test session.
The test session started on October 29, 2010. All tests were performed in Microsoft Windows 7 Ultimate x64 with all critical updates as of that date and the following drivers:
The graphics cards were tested in games in two resolutions: 1920x108 and 2560x1600. We believe that it doesn’t make sense to test these graphics accelerators in lower resolutions, because monitors supporting 1920x1080 are currently priced within very affordable $150-$170 range.
The tests were performed in two image quality modes: “High Quality+AF16x” – maximum texturing quality with enabled 16x anisotropic filtering and “High Quality+ AF16x+AA4(8)x” with enabled 16x anisotropic filtering and 4x full screen anti-aliasing (MSAA) or 8x if the average framerate was high enough for comfortable gaming experience. We enabled anisotropic filtering and full-screen anti-aliasing from the game settings or configuration files. If the corresponding options were missing, we changed these settings in the Control Panel of Catalyst and GeForce/ION drivers. Vertical sync was always off in driver control panels.
As usual, we have one more time updated our benchmarking games and applications list. Besides updating the current games with the latest patches, we added three new titles: Civilization V, F1 2010 and NBA 2K11. Moreover, we also added the latest campaign in the Left 4 Dead 2 sequel called “The Sacrifice”. So, the complete list of test applications included two popular semi-synthetic benchmarking suites, one technical demo and 19 games of various genres. Here is the complete list of tests used with the settings (all games listed in their release order):
If the game allowed recording the minimal fps readings, they were also added to the charts. We ran each game test or benchmark twice and took the best result for the diagrams, but only if the difference between them didn’t exceed 1%. If it did exceed 1%, we ran the tests at least one more time to achieve repeatability of results.
As we said at the beginning of this review, our aim is to carry out a comprehensive test session to find out how CrossFireX technology works with AMD’s new Radeon HD 6870 and HD 6850 cards as compared to Nvidia’s SLI configuration built out of two GeForce GTX 460s. Considering the positioning of the mainstream cards by the GPU developers, it would be logical to add a SLI tandem of two GeForce GTX 470s. But as we didn’t had them at the moment, we tried to make up for them by overclocking our GeForce GTX 460s from 675/1350/3600 MHz to 830/1660/4220 MHz. As our samples of the Radeon HD 6870 and 6850 cards differed greatly in terms of overclockability, we benchmarked them at their default frequencies in CrossFireX mode. We guess it is also going to be interesting to benchmark the Radeon HD 6850 at the frequencies of the HD 6870, i.e. 900/4200 MHz, and compare the results to those of the HD 6870. This will show us how the lack of 160 unified shader processors and 8 texture-mapping units affects the performance of the Radeon HD 6850.
In the diagrams below the results of the AMD Radeon HD 6870 1GB cards are marked in violet. The AMD Radeon HD 6850 and the CrossFireX configuration based on it are red; the GeForce GTX 460 is green (the Palit cards were benchmarked at the frequencies of the reference card). The individual cards and CrossFireX configurations are sorted in the diagrams in the order of descending retail price.
3DMark 2006 being too CPU-dependent when running on today’s top-end graphics cards, the diagram shows the results of the least CPU-dependent test HDR/SM3.0 rather than the overall scores (but you can find the latter in the table at the end of this part of our review).
Even though we had overclocked our six-core Intel Core i7 processor to 4.5 GHz, the CrossFireX and SLI configurations could not show their full potential when running 3DMark 2006 at the default settings. It is only at a resolution of 1920x1080 with full-screen antialiasing and anisotropic filtering that we can see the difference between the tested graphics solutions. The Radeon HD 6870 is the best of the single cards but the Radeon HD 6850 is a mere 2% behind when overclocked to the leader’s frequencies although the gap between the Radeon HD 6850 and HD 6870 is as large as 11% when both work at their reference clock rates. The GeForce GTX 460 proves to be the slowest single card here.
The multi-GPU technologies both boast highest efficiency in this semi-synthetic benchmark. To be specific, the Radeon HD 6870 tandem delivers a 93% performance boost and the Radeon HD 6850 tandem is 94% faster than the single card. The GeForce GTX 460 pair is as much as 98% faster than the single GTX 460! And that’s just the beginning! We can also note that the well-overclocked GeForce GTX 460s are still inferior to the Radeon HD 6870 pair working at their default clock rates.
The diagram shows the results of the GPU test which is the least CPU-dependent test from 3DMark Vantage.
Being a heavier benchmark than 3DMark 2006, the Vantage version allows each graphics card and multi-GPU tandem show their best even at the Performance settings. The GeForce GTX 460 feels more confident here against its opponents while the CrossFireX and SLI technologies are 90% and more efficient. The gap between the Radeon HD 6850 and the HD 6870 is larger here: 24-25% at the default frequencies and 9-11% when the 6850 is overclocked.
The first thing that strikes the eye in the Unigine Heaven Demo results is the impressive efficiency of CrossFireX technology. The Radeon HD 6870 and 6850 tandems outperform the respective single cards by 95% to 100% and more! Compared to that, the performance boost of 83% and 89% provided by the SLI configuration built out of two GeForce GTX 460s doesn’t look so impressive, even though it is quite large in itself. Moreover, the GeForce GTX 460 is in the lead here. The overclocked SLI tandem built out of two such cards is absolutely unrivalled. The overclocked Radeon HD 6850 is inferior to the reference Radeon HD 6870 by 4-5% which is very little considering the difference in their price.
Are you not impressed with the efficiency of CrossFireX technology in synthetic benchmarks? Here we’ve got the real game Crysis in which both multi-GPU tandems deliver a 100% performance boost relative to the single cards. The GeForce GTX 460 SLI works well here, too. However, its 83-89% efficiency is not as spectacular as those 100% or even more that you can get with the Radeon HD 6870 or 6850 tandems. It is only at the high-quality settings with 2560x1600 resolution that the multi-GPU configurations slow down, CrossFireX technology being 80% efficient and SLI, only 24% efficient. The GeForce GTX 460 is overall slower than its competitors here while the overclocked Radeon HD 6850, lagging behind the Radeon HD 6870 by 18-21% at the reference frequencies, overtakes the latter when overclocked to 900/4200 MHz.
The Radeon HD 6870 and HD 6850 pairs are as efficient as 100% and more in Far Cry 2, too. However, the GeForce GTX 460 SLI is not taken aback and performs 83 to 93% faster than the single card. As a result, the SLI tandem is competitive to the CrossFireX configurations based on the new solutions from AMD, especially at the overclocked frequencies and in the FSAA mode. The overclocked Radeon HD 6850 is almost as fast as the reference Radeon HD 6870.
Here is a third game in a row where both pairs of Radeons boast an impressive efficiency of CrossFireX technology. The GeForce GTX 460 SLI is less effective but falls behind its opponents in the FSAA-less mode only. The results are similar in the MSAA 8x mode although AMD's solutions are still somewhat faster.
Although we selected the most resource-consuming third scene out of the four test scenes available in Resident Evil 5, the efficiency of the Radeon HD 6870 (84-95%) is limited by the performance of the platform, particularly of the CPU. The two Radeon HD 6850s are slower and less dependent on the CPU performance, so they are 97-101% efficient compared to the single such card. The two GeForce GTX 460s in SLI mode work well, delivering a performance boost of 90 to 95%. When overclocked to the frequencies of its senior cousin, the Radeon HD 6850 is a mere 2-6% behind the Radeon HD 6870.
We’ve got very interesting results in this game. The two pairs of AMD Radeon cards are only 56-68% faster relative to the respective single cards. The GeForce GTX 460 SLI is far more efficient at 96-97%. The reason for such an unassuming performance of the new Radeons is in the driver because AMD’s graphics cards have always been as good as their Nvidia opponents in this game, also in multi-GPU configurations. Running a little ahead, we can say that this is not the only game in which the new Radeon cards are let down by their driver.
CrossFireX didn’t work smoothly in Borderlands before, so we can't expect anything exceptional from it now. Its efficiency varies from 40 to 50%. SLI technology is 93% efficient, helping the GeForce GTX 460 tandem win this test. The overclocked Radeon HD 6850 is a mere 3-5% behind the reference Radeon HD 6870.
Grand Theft Auto IV: Episodes from Liberty City is a third game in this test session where CrossFireX technology is less than 95-100% efficient. However, AMD's solutions are superior to Nvidia's in this game, our overclocking not helping the GeForce GTX 460 change this situation.
Oddly enough, CrossFireX proves to be incompatible with the latest game in the Left 4 Dead series. The CrossFireX tandems both fail in this test. On the other hand, the single Radeon HD 6850, let alone the HD 6870, is ahead of the more expensive GeForce GTX 460. By the way, the Radeon HD 6870 is 7-10% faster than the Radeon HD 6850 working at the same frequencies.
After a series of failures of CrossFireX technology, the Radeon HD 6870 and 6850 tandems improve. They deliver a performance growth of 90%, just like the GeForce GTX 460 SLI does. The GeForce GTX 460 and its SLI tandem is somewhere in between the HD 6850 and HD 6870 (and higher when it is overclocked). When overclocked, the Radeon HD 6850 gets closer to the HD 6870, the gap shrinking from 15-30% to 6-8%.
CrossFireX is highly efficient in this game, too. It improves the frame rate by over 90%. However, the results of the individual cards and multi-GPU tandems are not as high as we want. Metro 2033: The Last Refuge calls for more advanced graphics solutions or lower graphics quality settings. Overclocking makes the Radeon HD 6850 as fast as the Radeon HD 6870. Likewise, the overclocked GeForce GTX 460 SLI equals the Radeon HD 6870 CrossFireX.
Just Cause 2 is yet another game where CrossFireX technology is highly effective. The CrossFireX tandems enjoy an advantage of 95% and higher over the respective single cards. As for SLI technology, the GeForce GTX 460 tandem is 88 to 94% efficient, which is quite good, too. The Radeons are overall faster than the GeForce series products in Just Cause 2.
Aliens vs. Predator agrees with the previous game that CrossFireX technology is highly efficient: the Radeon tandems are 92 to 100% faster than the respective single cards. SLI technology is less effective at 86 to 92%. At the reference frequencies the GeForce GTX 460 is comparable to the Radeon HD 6850, the Radeon HD 6870 being somewhat ahead of them both. Aliens vs. Predator is only playable on these SLI and CrossFireX configurations at resolutions up to 1920x1080 if you turn on full-screen antialiasing.
CrossFireX is 100% effective again. This time we see such results in Lost Planet 2. SLI technology is good enough at 90% efficiency, but still inferior to CrossFireX. The overclocked GeForce GTX 460 SLI is ahead of the Radeon HD 6870 tandem.
The graphics subsystem is limited to 60 fps in this game, so the Radeon HD 6850 and HD 6870 can't show their best here. When it comes to the FSAA mode, these cards are limited by the game engine which is optimized for Nvidia products. AMD knows about that problem and made a public statement about tests in StarCraft 2: Wings of Liberty. We can only add that CrossFireX either doesn't work at all or only works at some specific settings here.
The Radeon HD 6870 and HD 6850 tandems are only 48 to 77% ahead of the respective single cards. While such results used to be viewed as excellent, we now expect higher efficiency considering what CrossFireX technology does in many other games. AMD’s programmers have something to do yet. The GeForce GTX 460 can offer a performance increase of 73-86% in SLI mode, which is quite typical of SLI technology.
Sid Meier’s Civilization V agrees with many of the previous tests: the Radeon HD 6870 and HD 6850 CrossFireX configurations are twice as fast as the respective single cards here. The GeForce GTX 460 SLI is somewhat less effective, being 90% faster than the single graphics card in Civilization V.
As a rule, multi-GPU technologies do not work effectively, if they work at all, in new games. But in the released F1 2010 CrossFireX technology is 97% efficient in the FSAA mode. Its efficiency drops to 43% in the FSAA-less mode, though. As for SLI technology, the GeForce GTX 460 tandem is 58 to 70% faster than the single card.
Multi-GPU technologies do not work at all in this new game. Well, NBA 2K11 doesn’t really need CrossFireX or SLI because the single cards deliver playable frame rates even at the highest resolution with FSAA. You don’t need more than one card to play basketball.
The two pairs of Radeon cards boast 100% efficiency of CrossFireX technology compared to the respective single cards (by the way, that's a tenth game in which they perform like that). Nvidia’s SLI technology is not so effective, yet the GeForce GTX 460 easily maintains its lead both when single and as the SLI configuration. Overclocking the Radeon HD 6850 to the frequencies of the Radeon HD 6870 reduces the gap between them from 13-17% to 3-5%.
Here is the table with full test results:
Now we can proceed to the performance summary diagrams.
The first two pairs of summary diagrams compare the efficiency of AMD CrossFireX and Nvidia SLI technologies (in percent of the performance of the respective single cards).
When used on the new AMD Radeon HD 6870 and HD 6850 graphics cards, CrossFireX technology is more effective than the SLI configuration built out of two GeForce GTX 460s. CrossFireX improves performance by 100% in as many as ten games! However, Nvidia's multi-GPU technology gains the upper hand in such games as S.T.A.L.K.E.R.: Call of Pripyat, Borderlands, Left 4 Dead 2: The Sacrifice, StarCraft 2: Wings of Liberty and Mafia 2. Neither technology worked correctly in NBA 2K11.
The next diagrams show the advantage of the Radeon HD 6870 over the Radeon HD 6850 at the default frequencies and when the latter is overclocked to the frequencies of the HD 6870 (900/4200 MHz).
The Radeon HD 6850 is an average 15-20% slower than its senior cousin across all the games and test modes but the gap shrinks to 3-6% when we overclock it. The overclocked Radeon HD 6850 is even as fast as the Radeon HD 6870 in certain games, resolutions and test modes.
And the last diagrams compare the performance of the AMD Radeon HD 6870 tandem in CrossFireX mode with the Nvidia GeForce GTX 460 SLI when the latter is overclocked to 830/4220 MHz (this is an attempt to emulate a GeForce GTX 470 SLI configuration). The performance of the CrossFireX configuration is shown relative to that of the SLI tandem:
We don’t see a clear winner here. The Radeon HD 6870 tandem fails in games where CrossFireX technology is not very efficient. Otherwise, it wins. You can see everything in the diagrams, though.
We measured the power consumption of systems with different graphics cards using a specially modified power supply. To create maximum load we launched FurMark 1.8.2 in stability check mode at 2560x1600 with 16x AF singly and together with Linpack x64 (LinX 0.6.4, 4750 MB, 5 threads). These two programs load heavily the graphics card and CPU, respectively, so we can determine the peak power draw of the whole system and see what power supply will suffice for it (taking the PSU’s efficiency into account). You can see the results in the diagram:
As you can see, the new Radeon HD 6850 and 6870 graphics cards are more energy efficient than the GeForce GTX 460. The computers with them consume less power irrespective of whether it is a single card or a multi-GPU configuration. The difference is not large, yet it is not in Nvidia’s favor. We can also note that each computer configuration with any graphics card has a dramatically lower power draw in idle mode. Judging by the results, none of the tested graphics cards needs more than 20 watts of power when idle.
We also measured the level of noise using an electronic noise-level meter CENTER-321 in a closed room about 20 sq. meters large. The noise-level meter was set on a tripod at a distance of 15 centimeters from the graphics card which was installed on an open testbed. The mainboard with the graphics card was placed at an edge of a desk on a foam-rubber tray.
The bottom limit of our noise-level meter is 29.8 dBA whereas the subjectively comfortable (not low, but comfortable) level of noise when measured from that distance is about 36 dBA. The speed of the graphics cards’ coolers was being adjusted by means of a controller that changed the supply voltage in steps of 0.5 V.
The results are shown in the next diagram:
Unfortunately, none of the graphics cards we test in this review can be called quiet or even comfortable in terms of noisiness. Each of them produces a lot of noise when you launch a 3D application. We were especially disappointed with the sound of the reference cooler of the Radeon HD 6850 series cards. It is shrill and has an irritating plastic resonance, so you can hardly stand it much longer than a few minutes. The Radeon HD 6870 was also a disappointment because it produced even more noise than the noisy reference cooler of the Radeon HD 5870 or 5830. The Palit GeForce GTX 460 Sonic Platinum seems to be the quietest graphics card among them, according to the graphs above, but we wouldn’t call it really quiet, either. So, if you are going to buy one of these graphics cards and you care about how noisy your computer is, you should be prepared to replace the reference cooling system.
The main fact we can infer from our today’s test session is that CrossFireX technology works more effectively on the new AMD Radeon HD 6870 and HD 6850 graphics cards than on the older Cypress GPUs and Cypress-based products. The performance growth from adding a second card into the graphics subsystem amounts to about 100% in ten out of the 19 games we performed our tests in! Such a high efficiency could only be achieved earlier in a couple of games that were specifically optimized for AMD’s architecture, but now you only have to add up the frame rates of two cards to find out the resulting performance of their CrossFireX configuration. We don’t claim that CrossFireX has become absolutely perfect. There are four or five games on our list in which CrossFireX is not so effective or does not work at all. However, this multi-GPU technology seems to deliver higher performance benefits than Nvidia’s SLI. We should also note that the Radeon HD 6870 and 6850 cards are new and their drivers are going to be optimized more and more in the next few months. We have no doubt the incompatibility of CrossFireX with Left 4 Dead 2 is going to be solved in near future.
Nvidia’s GeForce GTX 460 SLI configuration proves to be in between the CrossFireX tandems built out of two Radeon HD 6870s and two HD 6850s in terms of performance, being occasionally faster than both of them thanks to its polished-off drivers and the opportunity to choose the rendering mode manually. Thus, these three pairs of graphics cards are ranked up according to their recommended prices. When overclocked, the GeForce GTX 460 SLI 1GB is competitive not only to the Radeon HD 6850 but also to the senior HD 6870 model. The only downside we can note about the Nvidia card is that it has higher power consumption than its AMD counterparts.
We used off-the-shelf cards from Palit for this test session. The Palit GeForce GTX 460 Sonic Platinum has two drawbacks in our eyes. First, it comes with scanty accessories. And second, it is too noisy in 3D mode. And while we can put up with the first drawback, the second one is far more serious. The graphics card’s PCB was too compact for its developers to install a larger GPU heatsink and, for example, two rather than one fan, which would have kept the cooler efficient at a lower level of noise. Otherwise, the Palit GeForce GTX 460 Sonic Platinum is quite appealing with its pre-overclocked frequencies and modest price.
