by Alexey Stepin , Yaroslav Lyssenko
05/11/2011 | 12:39 PM
We’ve frequently criticized Nvidia for its predilection to developing exceedingly sophisticated single-processor graphics cards that often go too far ahead of their time. For example, the early versions of the G200 chip were manufactured on 65nm tech process that wasn’t appropriate for a graphics core incorporating as many as 1.5 billion transistors. As a result, the GeForce GTX 280 card had a TDP of over 200 watts and was measured to consume up to 178 watts in our own tests. It was only inferior to the dual-processor GeForce 9800 GX2 in this respect. Besides being hot and uneconomical, the G200 was also not very profitable to make on the tech process available for Nvidia at that time. The company eventually corrected this problem by releasing a 55nm version of the core under the codename of G200b.
Later, we had the same story with the GF100 which consisted of 3 billion transistors: the GeForce GTX 480 card consumed as much as 260 watts in 3D mode. It surpassed the power requirements of the Radeon HD 5970 which itself was far from economical.
Luckily for Nvidia, the situation improved with the introduction of 40nm tech process. While the GeForce GTX 480 could not lay claim to the title of the best single-processor graphics card of its time, the revised GF110 core made it possible for Nvidia to roll out the successful GeForce GTX 580. Thus, Nvidia’s tactical inferiority has proved to be a strategic victory. The company now has a premium-class product that can satisfy most gamers and is free from the typical downsides of multi-GPU solutions.
AMD hasn’t yet released anything like that. Its Radeon HD 6970 is obviously inferior to the GeForce GTX 580 whereas the Radeon HD 6990 is a dual-processor card and depends on software support to deliver its best performance. Nvidia’s GeForce GTX 590 is software-dependent as well, though. Generally speaking, single-card dual-processor products are largely the result of compromises the developer has to concede in terms of GPU configuration and clock rates to meet the size, temperature and electrical constraints. The resulting multi-GPU card is of course inferior to multi-card solutions, although faster than any other single graphics card.
Dual-processor graphics cards are very expensive and may cost you as much as $700 now. Their target audience is rather limited, especially as people who can afford them may prefer to build a graphics subsystem out of two or more top-end single-processor cards from AMD or Nvidia. So, our today’s review is for those users who are considering the option of buying a couple of GeForce GTX 580s and running them in a SLI tandem.
Let’s compare the specifications of the SLI tandem with those of top-end single- and dual-processor solutions available on the market.
As you can see, the GeForce GTX 580 SLI configuration differs from the GeForce GTX 590 in clock rates only, yet this difference is significant: 772/1544 MHz against 608/1215 MHz. It leads to an advantage in texture-mapping performance and fill rate. The SLI tandem is inferior to the Radeon HD 6990 in the former parameter, but the Cayman chip’s advanced texture-mapping subsystem often proves to be redundant in gaming tests anyway. The GeForce GTX 580 SLI specs are all about big numbers, including power consumption. When you see “2x244 watts”, you can understand why the clock rates of the GeForce GTX 590 are reduced to 608/1215 MHz.
Thus, this SLI tandem has only two downsides: price and bulkiness. However, these are not going to bother wealthy enthusiasts who assemble their systems in huge computer cases. The question is how faster the GeForce GTX 580 SLI is going to be in comparison with the GeForce GTX 590 and Radeon HD 6990 in games.
The GeForce GTX 590 and the Radeon HD 6990 are the only products that stand higher than the GeForce GTX 580 in the graphics card hierarchy. Therefore ASUS has tried to present its custom-made version of the latter card with as much pomp as possible. The large box doesn’t look very extravagant, though:
The green color denotes an Nvidia-based product. ASUS uses the same design but with a red background for graphics cards with AMD processors. There is a huge caption on the box reading: “DirectCU II – 20% Cooler”. As opposed to what we usually read on graphics card packages, this marketing claim is well-grounded because the card indeed features an extraordinary cooling system. The information about factory overclocking may be useful, too. Graphics card makers do not usually specify the exact clock rates of their products right on the box.
Inside the wrapper there is a black cardboard box with gold embossing. This type of packaging looks gorgeous but isn’t informative. The graphics card lies in a foam-rubber tray. There are boxes with accessories nearby:
Oddly enough, this is all we could find in the box with our ASUS ENGTX580 DirectCU II despite the price category this card belongs to. This is a bare minimum of accessories which doesn’t seem to match the exclusiveness of the product. Although you can use the card out of the box, you shouldn’t expect any nice extras like a free game or a cute CD case that used to be included with older ASUS products.
ASUS is one of the few graphics card makers that risk developing unique products with custom-made PCBs. The ENGTX580 DCII is but barely reminiscent of Nvidia’s reference design (on the right).
It is a truly monumental product as it takes as many as three slots in a system case. In fact, it is also desirable that a fourth slot be empty as well, so that the card’s cooling system could get the fresh air it needs. Compact system cases won’t do for this card because it is longer than the reference one. It is also incompatible with stiff SLI bridges. With the cooler removed, we can see the following:
The metallic back-plate, although equipped with a thermal pad, serves an aesthetic rather than a practical purpose. There is nothing to cool on the reverse side of the PCB because the power circuit elements are all on the face side. The only interesting thing we can spot here is the film capacitor NEC/Tokin 0E907. Notwithstanding its small size, it has a capacitance of 900 µF. ASUS suggests that this rare and expensive component is going to increase the overclocking potential of the ENGTX580 DCII. We’ll check this out shortly.
The face side of the PCB is a very exciting view because ASUS developers have left virtually nothing of Nvidia’s reference design. It is clear now why the card is so tall. The 10-phase power system wouldn’t fit onto an ordinary PCB with the given selection and layout of components. It’s somewhat odd that they used discrete components (power drivers and transistors) rather than integrated packs which would make the card smaller.
The power system employs high-quality capacitors and chokes that bear ASUS’s SAP logo. The SAP chokes have a solid core which eliminates the high-frequency noise of cheap ordinary chokes.
The graphics card’s power circuit features 8+2 design you can see on many top-end products. The GPU voltage regulator is based on a SHE ASP0907 chip. We saw it before on board the ASUS HD 6870 DirectCU and have reasons to think that it is a rebranded uP6208 or uP6225. The dual-phase memory voltage regulator is managed by an uP6223 controller.
Interestingly, the developers of this card have disabled overvoltage protection in order to give the user unlimited freedom to overclock. Although the large cooler should cope with any load you can throw at it, there is a higher risk of damaging a graphics card if it has no protection. You should keep this in mind when overclocking.
The ASUS ENGTX580 DirectCU II bears 12 GDDR5 memory chips from Samsung Semiconductor. These K4G10325FE-HC04 chips have a capacity of 1 Gb (32 Mb x 32) and a rated frequency of 1250 (5000) MHz, but they are clocked at 1002 (4008) MHz by the card, in full compliance with Nvidia’s official specs. The overclocking potential of the chips is limited by the 384-bit memory bus, so we do not expect to get to their rated frequency in our overclockability tests.
There is nothing extraordinary about the GPU. It is an ordinary GF110 revision A1 chip manufactured on the 46th week of 2010. It is a little bit pre-overclocked, from 772/1544 to 782/1564 MHz. ASUS suggests that further overclocking is up to the user. The GPU works in its full configuration with 512 ALUs, 64 texture-mapping units, and 48 raster back-ends. Despite its nonstandard PCB design, this card proved to be perfectly compatible with a reference GeForce GTX 580. Connected with the flexible bridge included with the ASUS card, they form the SLI tandem we are going to benchmark.
Our sample of ENGTX580 DCII turned out to have high overclocking potential. Its GPU would work at 824/1648 MHz without our resorting to extreme overclocking methods. The memory chips could not make it to their rated frequency and we had to stop at a memory clock rate of 1213 (4852) MHz. Overall, this is a good but not record-breaking result because some samples of GeForce GTX 570 can get as fast as 900 (1800) MHz in terms of their GPU frequency.
You can see the triple-tiered structure the ASUS card in the snapshot. As opposed to an ordinary GeForce GTX 580, the ENGTX580 DCII is additionally equipped with a DisplayPort which is becoming more and more popular. The card can be easily connected to any display device but does not support more than two displays simultaneously (you can make it three displays by enabling SLI mode).
The graphics card features an exclusive cooling system called DirectCU II. As its name indicates, the cooler implements the direct-contact technology which has become very popular among the manufacturers of top-end cooling systems.
The point of the mentioned technology is simple. The cooler’s heat pipes have direct contact with the GPU chip instead of using any intermediary materials. The technology works right only if implemented properly. Judging by the thermal grease imprint, the ENGTX580 DCII is such a case. Each of the five heat pipes serving the two sections of the large aluminum heatsink is put to work. There are two 92mm fans blowing at the two heatsink sections.
The power transistors of the voltage regulators are equipped with an additional small aluminum heatsink whereas the metallic back-plate cools them via an elastic thermal pad. Thus, the cooling system of the ASUS ENGTX580 DirectCU II is high quality but there was one problem with our sample of the card. The impeller of one fan was brushing against the edge of the cooler casing and we had to bend it a little to correct that.
It would be quite inappropriate for such an advanced cooler to work with much noise. The ENGTX580 DCII is indeed rather quiet and comparable to the GeForce GTX 570 in this respect in 3D applications. The speed of its fans, according to the tachometer, varies from 1200 to 1830 RPM. With the large heatsinks and the properly implemented direct-touch technology, there is no need for the fans to get faster unless at overclocking.
The temperature of the card is not very low, yet it is surely cooled more effectively than Nvidia’s reference sample in idle mode and as effectively as the latter in 3D mode. Overall, the cooling system of the ENGTX580 DCII is very good, but its large dimensions are a downside. The cooler requires at least two more slots besides the one occupied by the graphics card proper. And we’d even recommend leaving a third slot empty, so that the cooler’s fans could get fresh air. You should also keep it in mind that the DirectCU II cooler doesn’t exhaust the hot air out of the system case, so the latter must be properly ventilated.
We are going to investigate the gaming performance of GeForce GTX 580 SLI tandem using the following universal testbed:
We used the following ATI Catalyst and Nvidia GeForce drivers:
The ATI Catalyst and Nvidia GeForce graphics card drivers were configured in the following way:
Below is the list of games and test applications we used during this test session:
First-Person 3D Shooters
Third-Person 3D Shooters
Semi-synthetic and synthetic benchmarks
We selected the highest possible level of detail in each game. If the application supported tessellation, we enabled it for the test session.
For settings adjustment, we used standard tools provided by the game itself from the gaming menu. The games configuration files weren’t modified in any way, because the ordinary user doesn’t have to know how to do it. We ran our tests in the following resolutions: 1600x900, 1920x1080 and 2560x1600. Unless stated otherwise, everywhere, where it was possible we added MSAA 4x antialiasing to the standard anisotropic filtering 16x. We enabled antialiasing from the game’s menu. If this was not possible, we forced them using the appropriate driver settings of ATI Catalyst and Nvidia GeForce drivers.
Besides GeForce GTX 580 SLI tandem, we also tested the following products:
Performance was measured with the games’ own tools and the original demos were recorded if possible. We measured not only the average speed, but also the minimum speed of the cards where possible. Otherwise, the performance was measured manually with Fraps utility version 3.4.2. In the latter case we ran the test three times and took the average of the three for the performance charts.
The dual-card GeForce GTX 580 SLI configuration beats both single-card dual-GPU monsters in the first test, which is obviously due to the lower clock rates of the GeForce GTX 590. Anyway, the SLI tandem enjoys a 43% advantage over the latter at the popular resolution of 1920x1080 pixels. It is also 1 to 14% ahead of the Radeon HD 6990 which works in overdrive mode (you should keep it in mind that the overdrive mode is not the default one and is not covered by the warranty).
The top-end solutions are limited by the platform’s overall performance at 1600x900 but at the higher resolutions the dual-card SLI tandem is obviously superior to its single-card cousin: by 14% in the Full HD mode and by 22% at 2560x1600. The Radeon HD 6990 is unable to challenge the leader, either.
This game's system requirements being rather low, top-end solutions like GeForce GTX 580 SLI can't show their best in it. Anyway, Nvidia's dual-card SLI tandem is ahead at any resolution whereas the GeForce GTX 590 has certain problems at resolutions up to 1920x1080, even though its frame rate is never lower than 90 fps.
Crysis 2 is not as a resource-consuming application as the previous games in the series, yet the top-end SLI and CrossFire configurations are useful, especially at resolutions above 1920x1080 where other graphics solutions fail to deliver a playable frame rate. Nvidia's products are superior in the top-performance category although the GeForce GTX 590 isn’t much faster than the Radeon HD 6990. The GeForce GTX 580 SLI tandem proves that single-card dual-GPU solutions are always a compromise, outperforming the GeForce GTX 590 by 25% and the Radeon HD 6990 by 40%.
This game is tested with the tessellation option turned on.
The bottom frame rate is still very low, yet the GeForce GTX 580 SLI tandem is unrivalled in terms of average speed. It is 7 to 40% faster than its opponents, depending on the resolution. The gap from the Radeon HD 6990 clocked at 880 MHz is very small in the Ultra HD mode, though.
This game runs with enabled tessellation.
The GeForce GTX 580 SLI tandem is superior at every resolution. It is followed by the GeForce GTX 590 whereas the single GTX 580 is third up to 2560x1600. The advanced capabilities of the Fermi architecture in terms of geometry processing and tessellation show up here. Although the game is playable even on the modest Radeon HD 6970 in the Ultra HD mode, it’s comforting to have a reserve of speed here.
The Radeon HD 6990 is ahead at the low resolutions but Nvidia’s solutions overtake it at 1920x1080. The GeForce GTX 580 SLI then leaves the Radeon HD 6990 behind by 15%. The advantage of the SLI tandem over the rest of the tested products is that it ensures a higher bottom speed, up to 40 fps.
This game favors Nvidia and runs fast at 2560x1600 even on a GeForce GTX 580. Two such cards are unrivalled and beat the Radeon HD 6990 by as much as 50%! Of course, the GeForce GTX 580 SLI tandem is going to be more expensive than any dual-processor card but it offers unprecedented performance for that money. We guess there are enthusiasts who will be willing to pay $1000 for such a high speed.
We use a patch with high-res textures and FSAA.
Dragon Age II was used in AMD’s marketing materials and for a good reason. The Radeon HD 6990 is invincible in this test. The GeForce GTX 580 SLI tandem is the only Nvidia-based solution to deliver full comfort at 2560x1600, but it is 32-33% slower than the AMD-based flagship at 2560x1600 and 25% slower at 1920x1080.
We enforced full-screen antialiasing using the method described in our special Mass Effect 2 review.
This game isn’t new and its system requirements are not high. The top-end graphics subsystems can produce over 300 fps, which is overkill. Still, we can note that the GeForce GTX 580 SLI is 50 to 80% faster than the Radeon HD 6990 here.
The GeForce GTX 580 SLI enjoys but a very small advantage in this game: 5 to 14% at the first two resolutions. Then, at a resolution of 2560x1600, the Radeon HD 6990 goes ahead. Fortunately, Nvidia's solution delivers a playable frame rate in every mode.
Nvidia's cards have problems working together in SLI mode. Despite the higher clock rates, the SLI tandem is slower than the single GeForce GTX 590. The frame rate is playable even at 2560x1600 but investing into this graphics subsystem for playing StarCraft II doesn't make sense, especially as the Radeon HD 6990 offers you more speed.
This game makes you choose between FSAA and a full selection of special effects. We prefer the latter.
The GeForce GTX 580 SLI is most appropriate for playing Shogun 2, especially if you want to do that on an Ultra-HD monitor (2560x1600 or 2560x1440). The GeForce GTX 590 delivers good performance in the Ultra-HD mode too, yet the SLI tandem is about 25% faster!
We minimize the CPU’s influence by using the Extreme profile (1920x1200, 4x FSAA and anisotropic filtering). We also publish the results of the individual tests across all resolutions.
Our GeForce GTX 580 SLI configuration easily scores over 20,000 points. None of the single-card solutions can do that, even though the Radeon HD 6990 in overdrive mode and the GeForce GTX 590 should be given credit for almost reaching that mark. The dual-card SLI configuration is unrivalled in the individual tests except for the resolution of 2560x1600 in the second test where the Radeon HD 6990 working at 880 MHz turns out to be a little faster.
We use the Extreme profile here. As opposed to 3DMark Vantage, this profile uses a resolution of 1920x1080 pixels.
The GeForce GTX 580 SLI fails to score 4,000 points in this benchmark, but beats the GeForce GTX 590 by 25% and the Radeon HD 6990 (in overdrive mode) by 13%. We guess it will take a 3- or 4-way graphics subsystem to reach the next milestone.
This benchmark makes wide use of tessellation to render the surface of the earth. The number of polygons per one frame can be as high as 1.5 million!
The benefits of Nvidia’s architecture are not obvious at low resolutions in this test. However, the GeForce GTX 580 SLI tandem goes ahead at 2560x1600, outperforming the Radeon HD 6990 by 34% and the GeForce GTX 590 by 15%.
We use Normal tessellation in this test.
The results of this test at the highest resolution available on our platform prove the huge potential of the GeForce GTX 580 SLI configuration. It is the only graphics subsystem to deliver a playable bottom speed in the Ultra HD mode. Of course, you will hardly want to build such a configuration only for benchmarking it in synthetic tests, yet its performance in modern games is generally very high, too.
The GeForce GTX 580 SLI configuration offers an expectedly high level of performance. Costing about $1000, it is more expensive than the premium-class single cards such as GeForce GTX 590 and Radeon HD 6990, yet it is also generally faster than either of these products which come at a recommended price of about $700. With such price tags, the additional $300 can hardly matter much for an enthusiast who wants to have the fastest gaming computer. The unprecedented performance of the GeForce GTX 580 SLI subsystem is indicated by the summary diagrams:
Premium-class graphics subsystems are hardly ever used with inexpensive monitors which are limited to this display resolution. Anyway, we can note that the average advantage of the GeForce GTX 580 SLI over the Radeon HD 6990 is about as high as 29%. It is also 14% ahead of the GeForce GTX 590. SLI technology boasts superb scalability, the SLI tandem being up to 98% faster than the single GTX 580. It must be noted that the frame rate is often limited by factors other than the graphics subsystem performance when you play at this resolution.
At the resolution of 1920x1080, the average advantage of the SLI tandem over the GeForce GTX 590 increases to 16-17% but the Radeon HD 6990 gets somewhat closer (26%). As a matter of fact, there is no real need for the huge computing resources provided by the SLI pair of GeForce GTX 580s even at this resolution, perhaps with the exception of Aliens vs. Predator and Dragon Age II. Nvidia’s insistence on developing sophisticated single-processor graphics cards is finally justified because the single GeForce GTX 580 can satisfy 99%, if not 100%, of gamers at 1920x1080 which is the most popular display mode today.
Finally, we’ve reached the Ultra HD display mode which is the main battlefield for premium-class graphics subsystems. The GeForce GTX 580 SLI subsystem shows its best, beating the Radeon HD 6990 by 26% on average and by 60-80% in individual tests. The only exception is Dragon Age II which, incidentally, is used by AMD for promoting its graphics cards. The increased speed of the GeForce GTX 580 SLI can make a difference in terms of playability in such games as Aliens vs. Predator and Total War: Shogun 2. Otherwise, the GeForce GTX 590 should be quite enough.
The GeForce GTX 580 SLI has the typical downsides of any SLI configuration. First of all, it is bulky. In our case, it is especially bulky because one graphics card has a triple-slot cooler. There are not so many system cases and mainboards that are ready to accommodate it. Second, the GeForce GTX 580 SLI needs a lot of power. You have to double the power consumption of a single GeForce GTX 580, which is 265 watts, by two. The resulting number is very high. Moreover, you have to take care about proper ventilation to avoid overheat. Finally, this graphics subsystem is going to depend on software support on the side of the driver and game engine, but this problem is typical of single-card SLI-based solutions as well.
All in all, the GeForce GTX 580 SLI is a fast, expensive, hot and uneconomical toy for computer enthusiasts whose priority is speed. At the current moment, it is the fastest dual-GPU solution based on two individual graphics cards. If you want even more speed, you should consider 4-way configurations built out of two Radeon HD 6990s or GeForce GTX 590s.
As for the particular card we used for this review, the ASUS ENGTX580 DirectCU II is quite an exciting version of GeForce GTX 580 which features high-quality components, a custom-made PCB, and an advanced cooling system. It is targeted enthusiasts, especially at overclockers. It combines high quality with high performance. The single downside we can note about it is the large dimensions of its cooler. This card alone can occupy a lot of space inside your computer case, so you can hardly use a lot of other expansion cards alongside it.