by Sergey Lepilov
03/24/2011 | 06:00 AM
There is always a constant and tough struggle for supremacy among the manufacturers of computer components like CPUs, graphics cards, system memory modules, coolers, etc. There is fierce competition in each price category, especially among top-end products. The graphics card market is a vivid example of that. Having the world's fastest graphics card under one’s belt is not only prestigious but also profitable because it proves the manufacturer's technical superiority and promotes its sales in other price sectors.
Up to this moment the Nvidia GeForce GTX 580 has been the fastest single-GPU graphics card although AMD could offer its dual-processor Radeon HD 5970 as an alternative. On March 8, 2011, AMD released an even faster dual-GPU product, Radeon HD 6990. Nvidia hasn’t taken long to respond and has just rolled out its own dual-processor GeForce GTX 590. So, I will be testing these two new dual-GPU solutions in this review.
Let’s first take a look at the new cards’ specs in comparison with other solutions from AMD and Nvidia.
I've got reference samples of both cards without any packaging and accessories. So, let's take a look at them right now.
The AMD Radeon HD 6990 is as long as 305 millimeters, making it incompatible with quite a few system cases. Its height and depth are 101 and 40 millimeters, respectively. The Nvidia GeForce GTX 590 is smaller at 280x99x35 millimeters, so it can be installed into more system cases. Besides, the new card from Nvidia is somewhat lighter (1038 as opposed to 1139 grams). The face side of each card is covered with a plastic casing with a blower (HD 6990) or an axial fan (GTX 590).
The reverse side of the Radeon HD 6990 is covered with an aluminum plate except for the spots opposite the GPUs. The GeForce GTX 590, on the contrary, has such plates opposite the GPUs only whereas the middle part of the PCB is bare.
The cards differ in their connectivity options. The AMD Radeon HD 6990 has four mini-DisplayPorts and one DVI output whereas the Nvidia GeForce GTX 590 features three DVI outputs and one mini-DisplayPort.
There is a vent grid in each card’s mounting bracket for exhausting the hot air, yet some of that air remains inside the system case anyway.
Each card has two 8-pin power connectors at the top of the PCB.
They do not differ much in terms of the peak power draw: 375 watts for the Radeon HD 6990 and 365 watts for the GeForce GTX 590. AMD recommends a 750W or higher power supply with two 150W power connectors for its graphics cards. A 1200W or higher power supply is recommended for a CrossFireX tandem built out of two Radeon HD 6990s. Nvidia has the following recommendations: 700 and 1000-watt power supplies for a single GeForce GTX 590 and a SLI tandem, respectively.
Each card has a single connector for building multi-GPU configurations. It is located in the top front part of the PCB.
There is a small switch next to the CrossFireX connector on the Radeon HD 6990. Instead of switching between the main and backup BIOSes as is implemented in the Radeon HD 6970 and HD 6950, this switch selects the operation mode of the card's GPU. When turned on, it increases the GPU clock rate by 50 MHz. The GeForce GTX 590 has no such switch, but there are three vent slits in its casing next to the MIO connector.
With the coolers removed, we can see that the two cards have similar component layouts:
Nvidia’s dual-PCB sandwiches are already a long-forgotten solution, the GeForce GTX 590 being designed in the same way as dual-processor Radeons. AMD and Nvidia put both GPUs on a single PCB but at somewhat different locations which are determined by the placement of the graphics memory chips which are themselves installed on each side of the PCB. There are eight memory chips for each GPU. As you can see, each PCB is highly sophisticated and populated with lots of components.
The Radeon HD 6990 carries two full-featured Cayman GPUs. I want to stress the fact that they are indeed full-featured because dual-processor cards used to be equipped with cut-down versions of GPUs in the past. The GPU frequency of the Radeon HD 6990 is only 50 MHz lower than that of the Radeon HD 6970 and equals 830 MHz. However, there is a high-speed mode you can trigger by means of the abovementioned switch near the CrossFireX connector. The card's GPU frequency is 880 MHz in that mode, but AMD says that turning that switch on will make your warranty void. The card’s GPU frequency is lowered to 150 MHz in 2D applications to save power.
Each of its GPUs has 1536 unified shader processors, 96 texture-mapping units and 32 raster back-ends. As is typical of AMD cards, there are no protective caps (heat-spreaders) on the GPUs:
As opposed to AMD, Nvidia equips its GPUs with such caps. The company didn’t disable any subunits in the GPUs of its GeForce GTX 590, either. Each of the card's GPUs has 512 unified shader processors, 64 texture-mapping units and 48 raster operators. In other words, we've got two GeForce GTX 580 processors on a single PCB here. Their frequencies are lowered more than those of the AMD Radeon HD 6990, though. The GeForce GTX 590 clocks its GPUs at 607/1215 MHz, which is 21.4% lower than the clock rates of the GeForce GTX 580 (772/1544 MHz). The reason for this reduction is clear enough. If Nvidia used the clock rates of the GTX 580 for the GTX 590, the latter's heat dissipation and power consumption would be beyond all reasonable limits. The GeForce GTX 590 drops its GPU clock rates to 51/101 MHz in 2D mode as a power-saving measure.
Both cards have GDDR5 memory in 16 BGA-packaged chips. The Radeon HD 6990 carries a total of 4 gigabytes of graphics memory (2 gigabytes per each GPU) whereas the GeForce GTX 590 has 1.5 gigabytes of onboard memory for each GPU or 3 gigabytes in total. As usual, AMD installs Hynix H5GQ2H24MFR T2C chips on its Cayman-based reference cards. These chips have a voltage of 1.5 volts and a rated frequency of 5000 MHz. The card's memory frequency is 5000 MHz, too, but is lowered to 600 MHz in 2D mode. The memory bus is 256 bits wide.
The GeForce GTX 590 comes with Samsung K4G10325FE-HC04 chips that have a rated access time of 0.4 nanoseconds and a rated frequency of 5000 MHz. However, the card clocks them at 3414 MHz only, which is 15% lower than the memory frequency of the GeForce GTX 580 and 10% lower than that of the GTX 570. The memory frequency is lowered to 270 MHz in 2D mode. The memory bus is 384 bits wide.
This is what the GPU-Z tool reports about the two cards:
Unfortunately, version 0.5.1 does not recognize the GeForce GTX 590 and I had already given the card back by the time version 0.5.2 was released (March 21).
Now let’s have a closer look at the cards’ cooling systems:
The cooling systems of the Radeon HD 6990 and GeForce GTX 590 do not differ much in design, consisting of two heatsinks on a metallic base and a PWM-controlled fan in between them. The AMD card uses a classic 80mm blower whereas Nvidia prefers a 100mm axial fan.
The GPUs of the Radeon HD 6990 are cooled one by one: the air flows from the rear heatsink to the front one and is then exhausted. So, we can suppose that the GPUs are going to differ in temperature but most of the hot air will be exhausted out of the system case. The heatsinks of the GeForce GTX 590 are cooled equally because the fan drives the air towards both heatsinks, leaving half of the hot air inside the system case. The GPUs are likely to have similar temperatures as the result.
There are differences in the design of the metallic bases of the heatsinks. On the AMD card both heatsinks are soldered to it into a single whole whereas on the Nvidia card these components are separate, which makes it easy to assemble and dismantle the graphics card and, probably, ensures lower temperatures.
The heatsinks are hardly higher than 15 millimeters and look rather toy-like:
I wonder why Nvidia couldn’t make them some 5 millimeters higher. Both cards feature an evaporation chamber as you can see in the photos. Some dense gray-colored thermal interface is applied to the GPUs. Two types of thermal pads can be found between the cooler's base and the power components and memory chips.
Now let’s check out how hot these graphics cards are. To put them under load I ran the benchmark from Aliens vs. Predator (2010) in six cycles with maximum graphics quality settings at 2560x1600 with 16x AF but without FSAA:
It is at these settings that the GPUs had their highest temperatures. Monitoring wasn’t easy because the cards were so very new, but I got some information with MSI Afterburner 2.1.0 (by A. Nikolaichuk aka Unwinder) and with GPU-Z version 0.5.1.
Here are the results:
Both cards are very hot, just as expected. The temperatures of the Radeon HD 6990’s GPUs differ by 10-12°C whereas the difference between the GPUs of the GeForce GTX 590 is only 2°C. On the whole, the GeForce GTX 590 is colder than the Radeon HD 6990 by 10-12°C. Besides, I could measure the fan speed of the Radeon HD 6990 to be as high as 4800 RPM at such load. Of course, this was unbearably noisy. The GeForce GTX 590's fan is loud, too, yet its maximum speed of 2800 RPM is somewhat more comfortable.
At the same time, if Radeon HD 6990 runs at 830 MHz GPU frequency, the maximum graphics card temperature will drop by 5-6°C and the maximum fan rotation speed will not exceed 3100 RPM, which is, in fact, also pretty loud.
Now I want to add a few words about overclocking. The Radeon HD 6990’s cooling was too inefficient to hope for good overclocking, so I could only speed its GPUs up to 910 MHz. The memory frequency could be lifted up from 5000 to 6000 MHz, though, which is very good for a dual-GPU graphics card. As for the GeForce GTX 590, my sample could be overclocked from its default 607/1215/3414 to 725/1450/3940 MHz.
That’s good but hardly impressive. The Nvidia card’s GPUs didn’t get hotter when overclocked but the fan speed increased from 2800 to 3430 RPM. The Radeon HD 6990 got 2-4°C hotter when overclocked and its blower got even louder.
All graphics cards were benchmarked in a closed system case with the following configuration:
In order to lower the dependence of the graphics cards performance on the overall platform speed, I overclocked our 32 nm six-core CPU with the multiplier set at 24x and “Load-Line Calibration” (Level 2) enabled to 4.5GHz. 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 7-7-7-16_1T timings and 1.64V voltage. Turbo Boost and Hyper-Threading technologies were disabled during our test session.
Besides two graphics cards discussed above we also included two AMD Radeon HD 6970 graphics cards and a pair of Nvidia GeForce GTX 570 graphics cards working at the nominal frequencies:
The graphics cards were tested in CrossFireX and SLI tandem configurations as well as individually.
The test session started on March 19, 2011. All tests were performed in Microsoft Windows 7 Ultimate x64 with all critical updates as of that date and the following drivers:
We had to use different driver versions this time, because at the time of tests we didn’t have two Radeon HD 6970 and two GeForce GTX 570 graphics cards available to us at the same time when the drivers were released. Nevertheless, we performed a few additional tests aimed at comparing the performance of a single Radeon HD 6970 using Catalyst 11.2 and Catalyst 11.4 Early Preview drivers. The table below shows the obtained results:
Yes, there are a few performance changes, but they are not critical and should not affect the overall outcome.
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.
The benchmarking games and applications list hasn’t changed since our last review. It included two popular semi-synthetic benchmarking suites, one technical demo and 18 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 of the AMD Radeon HD 6990 are purple in the diagrams; the AMD Radeon HD 6970 is red, the Nvidia GeForce GTX 590 is blue, and the Nvidia GeForce GTX 570 is green. Due to time constraints the cards were only benchmarked at their default frequencies.
Since we already published a review of the Radeon HD 6990 in which we benchmarked the card at 830 as well as 880 MHz and found that these two modes did not differ much in terms of performance, the Radeon HD 6990 will be used in its accelerated mode here. This is relevant because some AMD partners still offer the warranty for the card even if it's used in the accelerated mode. We guess there will be more such vendors in the future.
The new dual-processor card from Nvidia is faster than its AMD counterpart in the FSAA-less mode and at low resolution. The difference is smaller in the high-quality mode at 1920x1080. At the same time, the dual-processor AMD flagship is as good as the Radeon HD 6970 tandem even though its memory frequency is lower by 500 MHz. The GeForce GTX 590, on its part, is somewhat slower than the SLI tandem of two GeForce GTX 570s. The new dual-processor card from Nvidia is clearly limited by its reduced clock rates and can't compete with the two GeForce GTX 570s, let alone GeForce GTX 580s.
The Radeon HD 6990 is the better of the two dual-processor cards in the new 3DMark 2011 although the gap from the GeForce GTX 590 isn’t large. Otherwise, we’ve got the same standings as in 3DMark Vantage.
This test is won by the GeForce GTX 590 which enjoys a 24% advantage in one of the test modes. The gap between the two dual-processor cards is smaller in the FSAA mode, yet Nvidia remains in the lead. As opposed to the two previous tests, the Radeon HD 6990 is somewhat slower than the Radeon HD 6970 tandem which boasts excellent scalability in CrossFireX mode in this test. We must also note that the GeForce GTX 590 is slower than the GeForce GTX 570 tandem, too.
The Radeon HD 6990 goes ahead in Crysis. The dual-processor graphics cards are both somewhat slower than the respective CrossFireX and SLI tandems.
The new dual-processor cards are close in performance in this rather old game. There is no practical difference between the frame rates of 144 and 148 fps, for example.
We see some fighting in this real-time strategy: the Radeon HD 6990 wins in the FSAA-less mode whereas the GeForce GTX 590 goes ahead when FSAA is turned on. AMD's dual-processor card is quite competitive to the CrossFireX tandem of two Radeon HD 6970s but Nvidia's new flagship is again inferior to the two GeForce GTX 570s in SLI mode (except for the high-quality mode at 2560x1600 where the two solutions deliver the same performance).
Resident Evil 5 seems to be the first game on my list where the green team is unrivalled. The GeForce GTX 590 is so good here that its performance in popular display modes is comparable to that of the two GeForce GTX 570s which have higher clock rates.
Nvidia is faster in S.T.A.L.K.E.R.: Call of Pripyat. AMD fans have to content themselves that the new Radeon HD 6990 is as fast as the two Radeon HD 6970s with higher clock rates.
The GeForce GTX 590 is again ahead of the Radeon HD 6990 which is an uncomfortably low bottom speed.
Besides noting that the dual-processor cards are similar in performance, we can also see that the multi-GPU technologies have low scalability in this game. I knew that from the previous test session, though.
Considering that each graphics solution delivers a very high frame rate in Left 4 Dead 2, the comparison has no practical worth but AMD fans can be pleased that the Radeon HD 6990 is as many as 23 fps ahead of the GeForce GTX 590 in the hardest display mode.
Today’s graphics cards are still not advanced enough to run Metro 2033: The Last Refuge at the highest graphics quality settings. This is indicated by the bottom frame rates which are the same for the single cards as well as multi-GPU configurations. The GeForce GTX 590 is ahead of its dual-processor opponent at 1920x1080 but the two are equals at 2560x1600.
The Radeon HD 6990 is just a little bit faster than the GeForce GTX 590 in Just Cause 2. There’s nothing interesting here otherwise. We can only note how effective both multi-GPU technologies are.
We’ve got the same picture as in the previous test.
The Nvidia GeForce GTX 590 is the faster of the two dual-GPU cards but still inferior to the pair of GeForce GTX 570s.
We’ve got a familiar picture in StarCraft 2: Wings of Liberty: the AMD cards are competitive in the FSAA-less modes but fail as soon as 8x FSAA is turned on.
The GeForce GTX 590 is indisputably faster in the easiest display mode (1920x1080 pixels without FSAA). The harder the settings, the closer the Radeon HD 6990 is to its opponent, finally overtaking the latter at 2560x1600 with FSAA.
The Radeon HD 6990 is somewhat faster than the GeForce GTX 590 but the overall picture is the same as in the other tests.
The 2011 season of Formula 1 is about to take off and the fans of this game series should know that the Radeon HD 6990 is faster than the GeForce GTX 590, yet both ensure a comfortable frame rate even at the highest settings.
The GeForce GTX 590 retaliates quickly as you can see in the diagrams. Nvidia wins this test, yet you should note that even a single Radeon HD 6970 is quite enough for playing Tom Clancy’s H.A.W.X. 2 with comfort.
There’s no point in commenting upon the results obtained in a beta version of a game. These numbers are not included into the summary diagrams.
Here is a table with the results of today’s tests.
Let’s move on to the summary diagrams.
The first pair of summary diagrams concerns with the main topic of this review. They help compare the AMD Radeon HD 6990 with the new dual-processor Nvidia GeForce GTX 590 card. Since AMD’s flagship came out earlier than Nvidia’s, the performance of the Radeon HD 6990 at a GPU clock rate of 880 MHz is the baseline. The performance of the GeForce GTX 590 is shown relative to it.
Neither card can boast overwhelming advantage over the other. The AMD Radeon HD 6990 is faster in 3DMark 2011, BattleForge: Lost Souls, Aliens vs. Predator (2010) and F1 2010, and also a little ahead of its opponent in Crysis, Far Cry 2, Left 4 Dead 2: The Sacrifice, Just Cause 2 and Sid Meier’s Civilization V. In its turn, the Nvidia GeForce GTX 590 is ahead in 3DMark Vantage, Unigine Heaven, Resident Evil 5, S.T.A.L.K.E.R.: Call of Pripyat, Borderlands, Lost Planet 2, Mafia 2, StarCraft 2: Wings of Liberty (at the high-quality settings) and in Tom Clancy’s H.A.W.X. 2. It also has a small advantage in Metro 2033: The Last Refuge. Thus, each card has 9 wins and the rest of the games/benchmarks are draws. If you want exact numbers, the GeForce GTX 590 is 1.6 to 6.8% ahead of the Radeon HD 6990 in the FSAA-less mode and 12.1 to 16.8% ahead in the FSAA mode. This is pure mathematics, though.
The next two pairs of summary diagrams show how faster the Radeon HD 6990 and the GeForce GTX 590 are against the Radeon HD 6970 and the GeForce GTX 570, respectively.
The dual-processor Radeon HD 6990 is an average 59-71% faster than the single-processor Radeon HD 6970 when we don’t use FSAA and 69 to 72% faster with FSAA. The GeForce GTX 590 is 52-66% and 66-78% faster than the GeForce GTX 570 in the FSAA-less and FSAA mode, respectively. You can see the details in the diagrams, so I can only add that the advantage of the dual-processor cards over their single-processor cousins depends on the efficiency of the CrossFireX and SLI technologies.
The last two pairs of summary diagrams compare the dual-processor Radeon HD 6990 with the CrossFireX configuration of two Radeon HD 6970s and the dual-processor GeForce GTX 590 with the SLI tandem of GeForce GTX 570s (the dual-card tandems are the baseline).
So, the dual-processor cards are overall close to the dual-card tandems in performance. And while I have no other questions about the Radeon HD 6990 and the Radeon HD 6970 CrossFireX, I wonder how faster a SLI configuration of two GeForce GTX 580s will be in comparison with the GeForce GTX 590. Considering the difference between the GTX 580 and GTX 570 and the efficiency of SLI technology, I suppose that a GeForce GTX 580 SLI tandem is going to add 5-7% to the performance of the GeForce GTX 570 SLI you can see in the diagrams.
It looks like neither of the two new dual-processor graphics cards can claim absolute superiority in games. The Radeon HD 6990 and the GeForce GTX 590 won the same number of my tests and delivered the same performance in the rest of them. We have this equilibrium for two simple but important reasons.
The first reason is that AMD improved its multi-GPU CrossFireX technology in the 6 series GPUs and graphics cards, so it can now often produce a near-100% increase in performance when you add a second Cayman or Barts-based card into your system. As a result, the Radeon HD 68xx and 69xx cards, being slower than the GeForce GTX 560/570/580 in single-card mode, are equal to or even faster than SLI tandems built out of those GeForces. Besides, the Radeon HD 6990 is actually equal to two Radeon HD 6970s and doesn’t differ from them in terms of hardware resources as was the case with AMD’s earlier dual-processor cards. All in all, AMD now has a product whose performance is comparable to that of Nvidia’s dual-processor flagship.
The second reason is that Nvidia seems to have played it too safe, fearing that its new card would be too hot and require too much power. While the GPUs have not been cut down in terms of functional subunits (they are identical to the GTX 580), their frequencies are set much lower and the resulting performance of the GeForce GTX 590 is not as impressive as might be expected. Perhaps Nvidia should have risked releasing this card with clock rates of 750/1500/4000 MHz for example, and this would leave no questions about who’s the leader now. On the other hand, we shouldn’t be upset about that because competition leads to lower prices. At the current moment, these new dual-GPU cards have the same recommended price of $699.
Comparing the new products from other aspects, the GeForce GTX 590 looks somewhat preferable to the Radeon HD 6990 because it is smaller, has a lower temperature of the GPUs (by 10-12°C) and a quieter cooler. But “quieter” doesn’t mean that it’s really quiet because each of these cards is rather noisy in 3D mode. I could not measure the power consumption of the cards for this review but this issue will be covered in our upcoming articles. Stay tuned.