by Sergey Lepilov
12/14/2010 | 09:07 PM
Just over a month ago AMD unveiled their new series of mainstream Radeon HD 6800 graphics cards based on Barts GPUs. And today, on December 15, 2010, they continue with their attack by releasing two top-end single-GPU solutions based on the Cayman graphics processor.
So far, there are only two cards in the series: Radeon HD 6950 and HD 6970. However, we won’t be surprised to see a third card with lower specs and a more affordable price, just as it was the case with the Radeon HD 5830. Moreover, a dual-processor flagship card based on the Cayman is scheduled for a Q1 2011 release as a replacement for the existing Radeon HD 5970.
AMD pits the Radeon HD 6950 and 6970 against Nvidia’s top-end GeForce GTX 570 and 580 cards. To make this even more exciting, the new products are expected to win not only through sheer performance but through competitive pricing and functional advantages. That’s an intriguing approach and we are going to discuss it in detail in this review.
Although the new Cayman processor is referred to as the HD 6xxx series, it is architecturally different from the HD 68xx as well as HD 5xxx solutions. The differences are both qualitative and quantitative, setting a milestone in the development of AMD’s new graphics architecture.
The most notable change is the transition from 5-way VLIW processors that debuted back in the R600 chip to 4-way VLIW ones.
The older 5-way VLIW processor contained five execution units, four of which could process only simple instructions such as addition and multiplication while the fifth one could also perform complex instructions like sine, cosine, square-root, etc. The new 4-way VLIW processor doesn’t have the complex-instruction unit as its functions have been transferred to the pair of simple units (numbers 3 and 4 in the slide). Thus, each 4-way VLIW processor is still capable of executing four simple or one complex instruction, which is the most typical scenario. It can only be inferior to the old 5-way VLIW processor in the rare situations when one complex and more than two simple instructions have to be performed simultaneously. This is unlikely in real-life applications whereas the higher efficiency of the new VLIW processors is going to increase the overall performance of the GPU at little cost.
The new GPU also features faster double-precision operations and an optimized mechanism of communication with system memory.
Another important improvement is the higher performance of the raster operators in different modes. The tessellation unit has also become much faster, according to AMD.
The rest of the innovations brought about by the Cayman processor are quantitative. The number of texture sampling and filtering units is increased from 80 to 96 while the number of VLIW processors is increased by 20%, from 320 (1600/5) in the Cypress to 384 (1536/4) in the Cayman.
Besides morphological antialiasing introduced in the Radeon HD 68xx series, the Radeon HD 69xx series features Enhanced Quality AA. An additional mode for MSAA, you can enable it in the Catalyst control panel:
Its principle is illustrated by the next picture:
Besides two, four or eight samples, there is also the same number of interpolated points which require less GPU resources to process. The image quality improves while the performance hit is promised by AMD to be no higher than 10% compared to the ordinary antialiasing algorithm.
The Cayman also has a new power-saving technology. You can manage it from the Overdrive tab of the Catalyst control panel:
You can find a new slider here which is called Power Control Settings. You can set it in a range from -20 to +20%.
This setting changes the limit for the graphics card’s peak power consumption. You can set it low to save power or high to achieve better overclocking results. The Cayman has a few sensors keeping track of its status and load that help make the new GPU more energy efficient. When these sensors identify a test program like FurMark, a protection mode will be triggered, lowering the frequency of the GPU. This is how it works when you launch the Perlin Noise test from 3DMark Vantage:
It is clear that the GPU clock rate is changing depending on the current load, but the frame rate remains at the same level. We doubt that this power-saving mechanism is going to be that effective in other applications, though. There should be a performance hit, even though people from AMD say that there are but few such applications. Well, we’ll see.
To push the limit of this official frequency throttling higher, you just move the slider to the maximum of +20%. Here is how this changes the graphics card’s behavior in the Perlin Noise test:
You can also do the opposite thing. That is, you can lower the card’s peak power consumption by setting that option at a negative value. This leads to a performance hit but AMD says you won’t lose much speed even in this case.
The power adjustment range is written into the graphics card’s BIOS and we hope there will be tools to tweak it (to enable higher values, in the first place). This would prevent your overclocking attempts to be limited by the frequency throttling mechanism.
Summing up this theoretical part of our review, we want to show you a couple of slides that compare the performance of AMD’s Radeon HD 6950 and 6970 with Nvidia’s GeForce GTX 470 and 480, respectively.
AMD also notes that, like with the Radeon HD 68xx series, the efficiency of CrossFireX technology is higher with the Radeon HD 69xx cards.
We will surely check this out in an upcoming review. Right now, let’s take a look at the specifications and see what the reference Radeon HD 6950 and HD 6970 cards are like.
The table below has graphics cards sorted according to their MSRP from highest to lowest (from left to right):
We’ve got two reference cards for our tests (non-reference Radeon HD 6950 and 6970 do not exist at this point, and may never come out at all). The only external difference between them is the sticker with model name on the cooler casing, and the power connectors. Each PCB is covered with the cooling system on the face side and with a decorative aluminum plate on the reverse side.
We call it “decorative” because it has no contact with any PCB elements while its protective properties are rather questionable. Each card is 276 millimeters long and 39 millimeters tall. The hot air from the cooler is partially exhausted through the grid in the top part of the casing, remaining inside the system case. It mostly goes out through the vents in the mounting grid, though.
There are as many as five video connectors there: two DVIs, two mini-DisplayPorts, and one HDMI.
There will be no special Eyefinity-ready Radeon HD 69xx cards. You will be able to connect up to six monitors to any of these cards using a special video hub.
Each model is equipped with two CrossFireX connectors for building multi-GPU configurations consisting of up to four graphics cards.
Next to them there is a small switch for choosing the BIOS the graphics cards will boot from. Yes, the reference Radeon HD 6950 and HD 6970 are equipped with two BIOS chips. One chip is programmable and another is a backup one to be used when the frequency or voltage settings in the first chip have been set too optimistically.
Now let’s see where the reference Radeon HD 6950 (hereinafter on the left) differs from the reference Radeon HD 6970 (hereinafter on the right). The first thing we can notice is the additional power connectors. Radeon HD 6950 has two 6-pin connectors while HD 6970 has one 6-pin and one 8-pin connector.
That’s okay since the senior model is specified to have a peak power draw of 250 watts and the junior - 200 watts. AMD introduces a new parameter, “typical gaming power”, for the Cayman. It is 190 watts for the Radeon HD 6970 and 140 watts for the Radeon HD 6950. The new cards have the same power draw of 20 watts in idle mode.
They also have nearly identical PCBs, save for a few components.
The power circuits are similar, too.
On closer inspection we can see some differences in the markings of the power circuit components.
The Cayman processors of the Radeon HD 6950 and HD 6970 cards are larger than the Barts and measure 389 sq. millimeters, but they still have no heat-spreaders. The GPU of our Radeon HD 6970 was manufactured 8 weeks later than the GPU of our Radeon HD 6950:
The GPU of the Radeon HD 6950 has 1408 unified shader processors and 88 TMUs. The GPU of the Radeon HD 6970 has 1536 unified shader processors and 96 TMUs. Each GPU has 32 raster operators. The default GPU frequency of the HD 6950 is 800 MHz and of the HD 6970 - 880 MHz.
In the power-saving mode the GPU frequency is lowered to 250 MHz and the GPU voltage is reduced from 1.1 to 0.9 volts on the HD 6950 and from 1.175 to 0.9 volts on the HD 6970. The rest of the GPU characteristics have been covered above in the theoretical part of our review and in the specifications table. We can only add that the diagonal distance between the mounting holes of the GPU heatsink is 75 millimeters on each card.
Each card comes with 2048 megabytes of GDDR5 memory in eight FBGA chips from Hynix Semiconductor. The Radeon HD 6950 carries H5GQ2H24MFR T2C chips which work at 1.5 volts voltage and 5000 MHz rated frequency. The Radeon HD 6970 is equipped with H5GQ2H24MFR R0C chips (1.5 volts, 6000 MHz).
As you could see in the specifications table, the Radeon HD 6950 supports 5000 MHz memory frequency while the Radeon HD 6970 clocks its memory at 5500 MHz. This leaves some elbowroom for memory overclocking. When idle, each card lowers its memory frequency to 600 MHz. The memory bus is 256 bits wide. All the memory chips are placed on the face sides of the PCBs.
The following screenshots sum up the characteristics of the new reference cards from AMD:
Each card sports a new dual-slot cooler that consists of a massive heatsink with evaporation chamber and slim aluminum fins, a metallic base, a blower, and a plastic casing on top of all that.
By the way, you can remove the casing by undoing the side locks, without taking the whole cooler off the card.
The evaporation chamber lies over the whole heatsink surface.
The chamber has no contact with the memory chips which are cooled by a metallic plate in the base of the cooling system. There are thermal pads in between.
Gray-colored thermal grease is used for the GPU.
The cooling systems are equipped with 80mm fans from Foxconn, PVB070G12N model. Unusually for graphics cards, the fan has a power rating of 24 watts.
The speed of the fan is regulated automatically by means of pulse-width modulation within 1200 to 5800 RPM interval. We can remind you that the top fan speed of the Radeon HD 58xx and 68xx series cards is no higher than 5000 RPM (and usually no higher than 4500 RPM), so we suspect that the new Radeons won’t be quieter than their predecessors. Let’s check this out right now.
To perform our temperature tests we installed each card into a closed system case which configuration is listed in the next part of the review. The ambient temperature was 25°C. As a typical load the system ran Aliens vs. Predator (2010) at 1920x1080 with 16x anisotropic filtering. As a high load we launched FurMark 1.8.2 from a renamed exe-file in Xtreme Burning Mode (at 2560x1600 and with 16x anisotropic filtering forced on in the Catalyst and GeForce/ION drivers). We used GPU-Z 0.4.9 and MSI Afterburner 2.1.0 beta 5 for monitoring.
First let’s see how hot the cards are under typical gaming load in the automatic fan control mode.
As far as Radeon HD 6950 graphics card is concerned, its thermal and acoustic performance are within acceptable limits, since the GPU only warmed up to 81°C and the cooling fan only sped up to 2260 RPM. At this point you can hear the fan working, but the overall noise from the graphics card remained with the acoustic comfort zone. As the number of runs in the gaming test increased, the temperature remained the same. Note that Aliens vs. Predator (2010) is one of the heaviest games in respect to graphics card thermal conditions.
Before we get to analyze the thermal readings taken off the Radeon HD 6970 graphics card, we have to say the following. The thing is that the AMD Radeon HD 6970 graphics card sample we received was not quite operational right from the start. It could work stably only in 2D mode, and as soon as we launched any 3D application, the fan rotation speed started jumping and so did the GPU temperature. We assumed that it was that notorious thermal throttling kicking in, therefore we changed the rotation speed adjustment limit to the maximum of +20%, but it didn’t amend the situation. So, assuming that the graphics card GPU was simply getting overheated, we removed the cooler and replaced the default dried out thermal interface with Arctic Cooling MX4. Only after this minor modification that graphics card started working stably in 3D applications. However, its thermal readings as well as the generated noise left much to be desired. For example, the GPU temperature rose as high as 104°C and the fan sped up to 4760 RPM! Of course, there is no comfort of any kind in this case. Luckily, Aliens vs. Predator (2010) turned out to be the only test in our list that heated up the graphics card so badly. However, we still tend to believe that the AMD Radeon HD 6970 sample we received wasn’t quite alright after all.
Nevertheless, we decided not to give up graphics card tests in FurMark, and here are the obtained results when the fans were working in automatic rotation speed mode:
And these are the results at maximum fan rotation speeds:
One more time we can hardly complain about the results obtained on Radeon HD 6950: 90 degrees C at only 2800 RPM in auto mode and 71°C at the maximum speed of 5770 RPM. As for Radeon HD 6970, it failed the FurMark test at maximum fan rotation speed and had to force the GPU frequency down to 500 MHz to complete.
Because of these shocking results shown by Radeon HD 6970 during our temperature tests and the absence of any alternative cooling solutions at the time (primarily because of the absence of VRM heatsinks), we decided to overclock only the Radeon HD 6950. Here is what we got:
The GPU frequency potential turned out pretty modest, since we managed to increase the clock speed by only 70 MHz (+8.8%), while the graphics memory really did well and hit the impressive 5880 MHz of effective frequency. I have to point out that even after overclocking the graphics card temperatures remained the same. In fact, it remained exactly the same. Only the fan started to work at 100-150 RPM higher speed depending on the type of load.
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 4.42GHz. The processor Vcore was increased to 1.425V in the mainboard BIOS:
The 6 GB of system DDR3 memory worked at 1.47 GHz frequency with 7-7-7-16_1T timings and 1.64V voltage. Turbo Boost and Hyper-Threading technologies were disabled during our test session.
We are going to compare AMD Radeon HD 6950 and HD 6970 against AMD Radeon HD 6870 1 GB and Inno3D GeForce GTX 570 1 GB:
Unfortunately, at this point we didn’t have enough time to include also GeForce GTX 580 into the current comparison, but it definitely will be in our upcoming articles, so stay tuned for more coverage.
The test session started on December 10, 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 currently most popular resolutions: 1920x1080 and 2560x1600. 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. Since the new 3DMark 2011 has been recently released, we removed the good old 3DMark 2006 from our test applications list. Besides, we also removed Colin McRae: DiRT 2 game and replaced it with a new game called F1 2010. We have also temporarily eliminated NBA 2K11 game, because we couldn’t find any logical explanation for the results obtained in it. Of course, we made sure that all the latest patches for the existing games were installed. So, the complete list of test applications included two popular semi-synthetic benchmarking suites, one technical demo and 17 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.
The results for AMD Radeon HD 6870 1 GB are marked purple, AMD Radeon HD 6950 2 GB in nominal and overclocked mode, as well as Radeon HD 6970 2 GB are marked red, while GeForce GTX 570 – green (we adjusted the working frequencies of our Inno3D graphics cards so that they would match the reference frequencies of Nvidia GeForce GTX 570). The graphics cards in the following diagrams are listed according to their MSRP (in descending order).
The new Radeon HD 6970 and HD 6950 look OK in this test, but only if compared against Radeon HD 6870, which falls about 40% behind them. They fail to beat GeForce GTX 570, so the “green” graphics card ends up ahead in both quality modes. Radeon HD 6970 is 11-15% faster than Radeon HD 6950 in this synthetic benchmark, but on overclocked mode the latter reduces the gap to only 3-5 %.
The newcomers look a little better in the new 3DMark 2011 test, where Radeon HD 6970 successfully competes against GeForce GTX 570, and HD 6950 follows closely behind. However, I am sure many of you have expected the single-GPU flagship product from AMD to win a much more convincing victory, because it is positioned as a direct competitor to GeForce GTX 580 but at a more attractive price. As for the advantage of the new Radeons 69xx over HD 6870, it makes 28%. The performance difference between the two new graphics cards is only 12% and gets even smaller when Radeon HD 6950 is overclocked.
Unigine Heaven is a benchmark that uses tessellation very actively, but it didn’t show anyone as a definite leader here. GeForce GTX 570 is ahead in 1280x1024 without antialiasing, while Radeon HD 6970 takes the lead in heavier graphics mode. The difference between Radeon HD 6970 and HD 6950 is 11% in favor of the senior model, but it reduces to 2-3% when the junior model is overclocked. Radeon HD 6970 shows the most impressive advantage over the HD 6870 model in this test: 59% in light image quality mode and 66% in heavy image quality mode with antialiasing enabled. Not bad at all!
Crysis remains one of the most resource-hungry games, where even top AMD and Nvidia graphics cards can only deliver comfortable gaming experience without antialiasing and in 1920x1080 maximum resolution. Nevertheless, GeForce GTX 570 runs neck and neck with Radeon HD 6970 when antialiasing is disabled, and as fast as Radeon HD 6950 – when antialiasing is enabled. The latter runs just as fast as its older sister when overclocked. The performance advantage of the new Radeon HD 6970 over HD 6870 makes between 24 and 56% depending on the resolution and image quality mode.
I have to admit that the new Radeon HD 69xx graphics cards do not look very convincing against the background of GeForce GTX 570 in Far Cry 2. Nvidia based graphics card is at least as good as the top AMD product when antialiasing is on, and outperforms the top AMD competitor (not to mention Radeon HD 6950) in low image quality modes. Overclocking helps Radeon HD 6950 to change the performance gap to HD 6970 from 7-12% to 1-2%.
Only in the third gaming test, BattleForge: Lost Souls, the new Radeon HD 6970 and HD 6950 proved to be faster than GeForce GTX 570, but only with antialiasing disabled. As soon as MSAA 4x is activated, Nvidia graphics card regains its lead. Radeon HD 6970 is about 10-11% faster than HD 6950, but this difference, however, is quickly eliminated when HD 6950 is overclocked. The top Cayman is about 25% faster than the top Barts solution.
GeForce GTX 570 is an indisputable leader here. Moreover, the new Radeon HD 69xx cards experience some performance issues in this game when antialiasing is on; they can’t even outperform the Radeon HD 6870.
The new Radeon HD 6970 appears to be the fastest of all today’s testing participants in S.T.A.L.K.E.R.: Call of Pripyat game. The second fastest is GeForce GTX 570, followed by Radeon HD 6950. At the same time I have to point out that the minimal fps rate on an Nvidia based graphics card is higher than on an AMD based one. Radeon HD 6970 is about 13-17% faster than HD 6950, but after overclocking of the latter card, this difference drops down to 3-4%. As for Radeon HD 6870, it falls 35-36% behind the new flagship product from AMD – Radeon HD 6970.
It is remarkable that when antialiasing is off, the new Radeon HD 69xx cannot catch up with Nvidia GeForce GTX 570. However, as soon as MSAA 8x is enabled in the drivers, the victory goes to Radeon HD 6970, while GTX 570 can only compete against the Radeon HD 6950 (which is exactly what AMD has initially planned). Radeon HD 6970 is about 10-12% faster than Radeon HD 6950 and about 22-26% faster than Radeon HD 6870 in three test modes out of four. However, in the heaviest test mode when the resolution is set at 2560x1600 and antialiasing is enabled, the new HD 6970 gets up to 82% faster than its predecessor, HD 6870. It most likely owes this advantage to twice as much memory that all Cayman based graphics cards have. And we are going to come across the same situation again today.
Here GeForce GTX 570 is the winner. All AMD based graphics accelerators demonstrate very similar results.
Keeping in mind that the fps rate in this not very resource-demanding, but very popular game is relatively high, antialiasing is totally justified. And in this mode Radeon HD 6970 outperforms GeForce GTX 570 and Radeon HD 6950 runs just as fast as the Nvidia based solution. The senior Cayman based product is about 12-13% faster than the junior one, which, however, disappears completely if the junior model is overclocked. I also have to add that Radeon HD 6970 is 24-30% faster than HD 6870.
This game didn’t reveal any leaders. All graphics cards are fairly slow here. I could only point out a decent performance gain of 24-36% that Radeon HD 6970 demonstrates over Radeon HD 6870.
In this game GeForce GTX 570 can successfully compete against the new Radeon HD 69xx only in one test mode out of four. When we increase the screen resolution or enable antialiasing, Nvidia based graphics card loses significantly not only to Radeon HD 6970, but also falls behind Radeon HD 6950. HD 6970 is 12-13% faster than HD 6950 in this game in nominal mode, but during overclocking it gets as close as 0-4%.
The picture here is similar to what we have just seen in Just Cause 2. Moreover, we notice an impressive performance increase when we move from Radeon HD 6870 to HD 6970: 33% without AA and 54-69% with MSAA 4x!
GeForce GTX 570 takes revenge after two losses in the preceding games: this time it leaves both new Radeons behind. Here I would also like to draw your attention to a significant performance difference between Radeon HD 6870 and Radeon HD 6970, when Cayman based product is 94% faster in the heaviest image quality mode and at least 31% faster in “lite” mode.
All AMD graphics cards experience some performance issues with enabled antialiasing in StarCraft 2: Wings of Liberty game. While without antialiasing all Radeon graphics cards are faster than GeForce GTX 570, then with MSAA 8x forced in the drivers they simply crash down showing extremely low fps rates. at the same time, Nvidia based solution doesn’t have any problems like that in this game.
Radeon HD 6950 successfully opposes GeForce GTX 570, while the flagship solution, HD 6970, wins in three test modes out of four. Although the overall performance difference between these three is not that dramatic.
The results obtained in this game stand out among others, but they are particularly remarkable due to the fact that in 2560x1600 resolution GeForce GTX 570 is defeated by both new products from AMD, while in less resource-demanding 1920x1080 resolution we don’t see anything like that. Could Nvidia graphics cards really lack video memory in the highest resolution?
GeForce GTX 570 competes against Radeon HD 6950 in the new F1 2010 game, while Radeon HD 6970 is slightly ahead of both of them in all modes except for the one with lowest image quality. Overclocking helps Radeon HD 6950 almost completely eliminate the performance difference between them. The higher is the video memory utilization and the need for large amounts of it, the farther behind falls Radeon HD 6870. For example, the advantage of Radeon HD 6970 over HD 6870 hits the impressive 64% in 2560x1600 with 8x antialiasing enabled.
I believe there is nothing to comment on here: Nvidia is an indisputable leader in this test.
In conclusion I would like to offer you a complete results table. So, now we can continue to the summary diagrams.
The first two pairs of summary diagrams should help you estimate the advantage of AMD Radeon HD 6970 2 GB over HD 6950 2 GB in nominal operational mode and during overclocking of the latter to 870/5880 MHz frequencies. The zero axis is the performance of Radeon HD 6950 and the performance of the flagship HD 6970 product is shown in percents from the latter:
Radeon HD 6970 is on average about 10-11% faster than Radeon HD 6950 without antialiasing and about 11-13% faster with enabled AA. It is remarkable that this graphics card reaches its highest (20%) and lowest (2%) performance advantage in the same game: StarCraft 2: Wings of Liberty. The reason for that is fairly simple: AMD Radeon graphics cards behave inadequately when antialiasing is enabled. Overclocking helps Radeon HD 6950 almost eliminate the performance difference from HD 6970, and in some games and test modes even outperform the latter.
The next set of summary diagrams compares nominal Radeon HD 6970 and HD 6950 against GeForce GTX 570. The results of the latter serve as a zero axis, with the gain or lag of the new competitors shown as percentage from this axis:
This battle ends with variable success on both ends. Since there is very small performance difference between Radeon HD 6970 and HD 6950, GeForce GTX 570 manages to successfully compete against both of them. We see parity between AMD and Nvidia or a tiny advantage of either one of them in quite a few games. In some games like Tom Clancy's H.A.W.X.2, Lost Planet 2, Borderlands and high image quality modes in StarCraft 2: Wings of Liberty GeForce GTX 570 is an obvious winner. In other games, Radeon HD 69xx is faster. It is hard to draw a definite conclusion here, because besides the gaming engine we should also keep in mind image quality settings and screen resolutions.
Finally, the third pair of diagrams is dedicated to comparing Radeon HD 6970 against Radeon HD 6870, both being top graphics accelerators in their class. The performance advantage of the single-processor AMD solution is shown in percents from the performance of HD 6870:
Looks like Radeon HD 6970 outperforms Radeon HD 6870 on average by 26-31% without antialiasing, and by 27-45% with antialiasing. The very last diagram shows the benefits of having 2 GB of video memory onboard: in some games the performance difference between the two cards is truly impressive.
We measured the power consumption of systems with different graphics cards (the monitor excluded) using a specially modified power supply at two loads. The typical gaming load was emulated by running two cycles of the Aliens vs. Predator benchmark at 2560x1600 with 16x AF and 4x MSAA. 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:
The power consumption of the graphics cards is proportional to their recommended prices. The more expensive a product is, the more power it demands. Interestingly, the difference between the Radeon HD 6970 and HD 6950 in peak power consumption is about 100 watts in the gaming mode although it should be only 50 watts according to AMD. On the other hand, we measure the full consumption of the whole computer, so some inaccuracies are possible.
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 adjusted with a controller that changed the supply voltage in steps of 0.5 V.
The results are shown in the next diagram:
The vertical lines show the fan speed range in the automatic mode when the cards were running FurMark: red dotted lines for the Radeon HD 6950, red solid lines for the Radeon HD 6970 and green lines for the GeForce GTX 570.
As you can see, the reference cooler of the GeForce GTX 570 is the quietest one throughout the entire speed range, but its speed is lower, notwithstanding FurMark. Its max speed was only 2430 RPM after 12 minutes of testing. The new Radeon HD 69xx series cards are not so quiet. The Radeon HD 6950 had a fan speed of 2820 RPM and the Radeon HD 6970, as high as 4760 RPM. But as we wrote above, our sample of the latter card seems to have some problems, so we wouldn’t be so sure about the noise level of the reference HD 6970. We can also note that each card was much more comfortable in terms of noisiness under gaming load.
Although AMD's Radeon HD 6950 and Radeon HD 6970 are not absolutely superior to the Nvidia GeForce GTX 570 in sheer speed and cannot match the GeForce GTX 580, except for specific games and benchmarks, these new graphics cards are quite appealing offers due to their lower recommended prices. The Radeon HD 6950 is $50 cheaper than the GeForce GTX 570 whereas the Radeon HD 6970 is $130 cheaper than the GeForce GTX 580. We must note, however, that the actual retail prices may differ from the recommended ones. Besides, both top-end cards from Nvidia are already available whereas the new Radeon HD 69xx series is yet to come to retail stores.
The Radeon HD 6950 is 10-13% slower than the Radeon HD 6970, but this gap can be easily closed by overclocking. Considering the $70 difference in price and the same amount of graphics memory (2 GB), the junior model seems even more appealing than the senior one. Besides, the Radeon HD 6950 has lower temperatures and, consequently, produces less noise, even though we can't call it cold and quiet. Users preferring silent computers should think about replacing the reference cooler, perhaps by installing a water-block. We like that the new graphics cards have a backup BIOS chip, so you won't need a second card to restore the main BIOS chip if it gets corrupt. The new Enhanced Quality AA mode and the power-saving technologies will be discussed in detail in our upcoming reviews.
What's next? Well, we are now waiting for Nvidia's response. Hopefully, it will involve a reduction in GeForce GTX 570 and 580 prices and a release of a GeForce GTX 560 model with attractive price/performance ratio. Tough market competition is always good for end-users, and that's why the Caymans are good, too.