by Alexey Stepin , Yaroslav Lyssenko
08/10/2009 | 10:04 AM
Gamers are always interested in how ultra-fast graphics solutions perform even though only few of them eventually buy such products. Catering to our readers’ demands, we have already published two reviews of top-end multi-GPU graphics subsystems. The first review compared ATI’s Radeon HD 4890 CrossFireX with Nvidia’s GeForce GTX 285 SLI. We learned that AMD/ATI’s solution, despite its lower price, is competitive to the pair of more expensive cards from Nvidia in many games and, oddly enough, at high resolutions (1920x1200 and higher).
However, today’s multi-GPU subsystems come in a variety of shapes and are not limited to two GPUs only. As soon as we had acquired a mainboard supporting three Nvidia cards in SLI mode, we pitted against each other 3-way GeForce GTX 285 SLI, 3-way Radeon HD 4890 CrossFireX and 4-way Radeon HD 4870 X2 CrossFireX configurations (see this review for details). The 3-way GeForce GTX 285 SLI came out the overall winner of that test session.
Nvidia also offers a 4-way multi-GPU solution called Quad SLI. This platform has had a long and unlucky history. Its first generation, announced in 2006, made use of specially developed dual-processor GeForce 7900 GX2 cards, and our tests proved that it brought in more problems than benefits. The platform was highly unstable, showed various image artifacts and was occasionally inferior to dual-processor Radeon HD 1900 XT CrossFire and GeForce 7900 GTX SLI in sheer speed. A Quad SLI system based on widely available GeForce 7950 GX2 was free from most of those shortcomings but proved to be slower than a Radeon HD 1950 XTX CrossFire tandem in 12 out of the 20 tests. It was no triumph for Nvidia.
The second-generation Quad SLI, discussed in our There Will Be Speed: The Clash of Modern Multi-GPU Technologies article, was based on unified architecture, but the pair of GeForce 9800 GX2 cards was still unable to win. A pair of Radeon HD 3870 X2 was faster in many games, especially at high resolutions. The lack of multi-monitor configuration support was a downside, too. Such configurations are quite popular among PC enthusiasts.
The release of the GeForce GTX 295 had naturally led to the third generation Quad SLI, but the graphics card itself proved to be too costly. It was actually as costly to make as to be unprofitable for Nvidia. The card did not really take off among wealthy gamers. Its availability was low and we did not have the opportunity to benchmark a configuration with two GeForce GTX 295 cards. The release of the new, single-PCB version of GeForce GTX 295 we have reviewed recently changed this. The simplified design and reduced manufacturing cost can make it popular. Many of Nvidia’s partners have released such products, some of which have already made it to our labs. Thus, we are now able to benchmark a GeForce GTX 295 SLI tandem to ultimately answer the question what multi-GPU subsystem is the fastest of all.
The Quad SLI subsystem will be assembled out of two identical cards: Gainward GTX 295 1792MB and Gigabyte GV-N295-18I-B. Let’s take a closer look at them first.
The Gainward GTX 295 1792MB comes to retail in a large black-and-blue box. Perhaps somewhat gloomy, this design looks restrained and quite appealing. There is a picture of an angel with outspread wings on the face side of the box.
Besides the angel, there are two common mistakes here: the memory type is indicated as DDR3 instead of GDDR3 and the memory amount is declared to be 1792MB although 3D applications can only make use of half that amount, i.e. 896MB. The package design is unified, so the features and specs of the specific model are indicated with a sticker.
The packaging has good protective properties. The graphics card lies on a foam-rubber tray covered by a sheet of the same material. Besides the card, the box contains the following:
That’s not much, but the lack of power adapters is hardly a serious problem since the Gainward GTX 295 is going to be installed into new systems. On the other hand, a 6-pin → 8-pin power adapter would be appropriate because not all PSUs provide an 8-pin graphics card connector.
Besides the GeForce 185.85 driver, the enclosed disc contains a full version of the user manual, Badaboom (a video transcoder that can utilize GPU resources), vReveal (a video enhancement tool that can use GPU resources, too) and the exclusive Gainward EXPERTool. The latter can be used for overclocking and to control the speed of the cooler’s fan instead of the traditional RivaTuner.
We don’t think the accessories to the Gainward GTX 295 match the price category the product belongs to. A copy of a popular game would be appropriate in this kit. Power adapters would make useful accessories, too. We tested a presale version of the product, however. The retail version may prove to be better in this respect.
This product is packed into a smaller and upright box as is typical of Gigabyte’s graphics cards. The box is painted a calm mix of white and blue. Despite the use of the metallic effect, the coloring does not irritate the eye. The box is embellished with a picture of a pretty girl with a huge gun.
There is no technical info on the box other than the model name, the type and amount of graphics memory. The memory type is indicated correctly as GDDR3 but the manufacturer couldn’t help attracting the customer with a larger but misleading number: 1792MB instead of 896MB that 3D application can actually use.
Inside the thin glossy wrapper there is one more box, made from robust black cardboard. The graphics card is fixed securely in a foam-rubber insert that protects the device against any hazards during storage and transportation. Accessories can be found nearby:
There are no special extras and the kit looks modest overall. Unlike with the above-described Gainward, you won’t have problems connecting power to the card even if your PSU has no PCIe connectors at all.
The enclosed disc contains the GeForce 186.18 driver. There are no additional software save for Yahoo! Companion. Thus, the accessories to the Gigabyte GV-N295 are better than those of the Gainward GTX 295 in terms of hardware but worse in terms of software. Anyway, we guess both kits don’t look nearly as good as the Inno3D GeForce GTX 295 Platinum kit we have reviewed recently as the latter contains as many as two high-quality games. Generally speaking, we don’t think it is right for expensive graphics cards to come with such scanty accessories.
We gave you a detailed description of the new GeForce GTX 295 in our Inno3D GeForce GTX 295 Platinum review. These two graphics cards also share Nvidia’s reference design, so there is no need for us to describe them. You can just refer to the mentioned review. The cards only differ with the picture on the cooler casing. Our sample of the Gainward card has no sticker at all, but the retail version will have a sticker with the same picture as on the card’s box.
The new GeForce GTX 295 resembles the GeForce 7900 GTX externally due to the characteristic design of the cooler with a central 80mm fan. The plastic casing is secured with locks and can be easily removed. Then you can see that the cooler actually consists of two separate heatsinks, one for each GPU.
Following the latest trends, each heatsink has a copper base that contacts with the GPU surface and is connected with heat pipes to the aluminum heatsink consisting of thin aluminum plates. The hot air from the left heatsink is exhausted out of the system case whereas the hot air from the right heatsink goes in the opposite direction, right into the interior of the computer, and may worsen the temperature of the other system components.
If you take off the heatsinks and remove the cooler’s base, you will see the following:
The power circuit of the new GeForce GTX 295 consists of two 3-phase voltage regulators based on Renesas R2J20651NP integrated packs and controlled by Analog Devices ADP3192A chips. Two single-phase voltage regulators are responsible for the memory chips. These regulators are based on integrated packs that combine power MOSFETs and drivers.
Power is connected via 6- and 8-pin connectors. The 8-pin power cable must be plugged in. Otherwise the card won’t start, reporting a power problem with the LED on its mounting bracket. This protection can still be easily bypassed by closing the two outermost pins in the 8-pin connector and plugging a PCIe 1.0 power cable in. In this case, it is necessary that the PSU had a sufficiently high load capacity of the +12V power rail. Nvidia recommends using 680W or higher PSUs, but this recommendation seems to be an overstatement for a single GeForce GTX 295.
The GPUs and memory chips of the graphics cards from Gainward and Gigabyte are clocked at the frequencies of the reference sample, i.e. 576/1242MHz for the GPU and 1008 (2016) MHz for the memory. Each GPU is equipped with a dedicated 896MB bank of GDDR3, and it is 896 megabytes of memory that 3D applications can access due to the specifics of today’s multi-GPU technologies. The memory buses are 448 bits wide.
We tried to overclock both cards and achieved the following results:
Gainward GTX 295 1792 MB
Gigabyte GV-N295-18I-B
The Gigabyte is somewhat better at overclocking, but the difference is negligible. The frequency growth is higher with both cards than with the Inno3D GeForce GTX 295 Platinum.
Each card is equipped with two DVI-I ports, one S/PDIF input and a MIO connector. We need the latter as we want to check out two GeForce GTX 295 in SLI mode. Before we do that, we will check out the noise and power consumption of such a tandem.
We discussed the noisiness of top-end multi-GPU systems in our comparative review of Nvidia GeForce GTX 285 3-way SLI and ATI Radeon HD 4870 4-way CrossFireX configurations. And now we will measure the noise produced by an Nvidia Quad SLI platform. Here are the results:
Although the new GeForce GTX 295 is not very noisy, two such cards get quite loud after they have worked long in 3D mode. The noise is quite uncomfortable then. This is no wonder as the bottom card is breathing freely but the top one is placed too close to the other and has problems getting enough of fresh air. The old version of GeForce GTX 295 could get some air through the window in the top PCB, but the new version doesn’t have this opportunity and has to increase its fan speed to avoid overheat. Fortunately, we had no stability issues like with the GeForce GTX 285 3-way SLI, and the tandem passed all of our tests. Still, we want to remind you that a multi-GPU subsystem requires a roomy system case with good ventilation. A system fan on a side panel would help the graphics cards breathe more easily.
Our power consumption testbed cannot test more than one graphics card, so the numbers below are just the total of the results of the respective individual cards in 3D mode.
The real numbers are going to be somewhat different because the power consumption of the Slave card is not zero even in 2D mode, but you can get a general notion of the power consumption of top-end multi-GPU solutions.
A GeForce GTX 295 SLI tandem is going to be more economical than a system with three GeForce GTX 285 cards, but the peak power draw of 400W is not low, even though it is indeed lower than that of the 4-way solution from AMD/ATI. A hypothetical Radeon HD 4890 4-way CrossFireX subsystem would consume over 450W, too.
Again, building a top-end multi-GPU solution with three or four graphics processors is not only expensive but also a noisy and uneconomical business. You should not expect such a subsystem to have a modest power draw. So, if you go in for it, you should purchase a roomy system case as well as a high-quality 750W or higher power supply.
We are going to investigate the performance of Nvidia Quad SLI graphics configuration using the following testbed:
The graphics card drivers were configured in the following way:
ATI Catalyst:
Nvidia GeForce:
The list of benchmarks includes the following gaming titles and synthetic tests:
First-Person 3D Shooters
Third-Person 3D Shooters
RPG
Simulators
Strategies
Semi-synthetic Benchmarks
We selected the highest possible level of detail in each game using 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 made a few exceptions for selected games if that was necessary. We are going to specifically dwell on each exception like that later on in our article.
Besides GeForce GTX 295 SLI tandem we have also included the following graphics accelerators to participate in our test session:
We ran our tests in the following resolutions: 1280x1024, 1680x1050, 1920x1200 and 2560x1600. 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.
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 2.9.8. In the latter case we ran the test three times and took the average of the three for the performance charts.
Starting from version 1.3 we use the game’s integrated benchmarking options together with a custom demo record. Unfortunately, this method does not report the bottom frame rate.
The Quad SLI platform’s advantage over the single GeForce GTX 295 is growing up along with the display resolution from 19 to 47%, yet even this 47% performance increase can hardly justify the increased amount of noise and high power consumption as well as the financial investment into the second GeForce GTX 295. Besides, the system with three GeForce GTX 285 is faster. AMD’s solutions with two Radeon HD 4870 X2 or three Radeon HD 4890 feel far more confident than the pair of GeForce GTX 295 cards. That’s not an encouraging beginning, really.
The game uses deferred rendering and does not support FSAA. Instead, it uses integrated antialiasing algorithms that make use of the GPU’s computing resources.
The GeForce GTX 295 tandem has problems in this game. The two extra GPUs provoke a twofold performance hit relative to the single such card. The 3-way SLI behaves in the same way, losing to the ordinary GeForce GTX 285 SLI pair. AMD’s solutions feel much better. The Radeon HD 4870 X2 tandem delivers an excellent performance at 2560x1600.
The GeForce GTX 295 SLI tandem cannot deliver a playable frame rate at 2560x1600 and is 1 to 9% slower than the 3-way GeForce GTX 285 configuration. The Quad SLI platform shows good scalability, though: it is 51% faster than the single card in terms of average speed.
We disabled the integrated frame rate limiter in the game console for the sake of comparing the cards. The game’s built-in benchmarking options do not provide information about the bottom speed, so there is no such info in the diagrams.
The top-end graphics solutions are overkill for this game due to its integrated frame rate limiter. However, we can see that if the limitation is disabled, the GeForce GTX 295 SLI tandem is slower than the GeForce GTX 285 3-way SLI by 1.5% to 10% depending on the display resolution. Thus, the Quad SLI subsystem still doesn’t look victorious.
The Quad SLI platform takes the lead in Far Cry 2, but its gap from the GeForce GTX 285 3-way SLI is less than 1.5% at 2560x1600 and its bottom speed is even lower than that of the 3-way platform. The scalability is far from impressive, varying from 14 to 39%. This can hardly be worth the money (over $1000) you have to pay for Quad SLI.
The system with two GeForce GTX 295 cards has the worst results among the multi-GPU solutions, its scalability being zero at low resolutions. At a resolution of 1920x1200 there is a negligible growth of average performance, not even 10% large. It is only at 2560x1600 that the quad-GPU platform from Nvidia shows its best and goes 59% ahead of the single graphics card. Anyway, it is the GeForce GTX 285 3-way SLI that wins this test.
The game runs on the Source engine and has an integrated benchmark, but the latter does not report the bottom speed information.
It is only at the resolution of 2560x1600 that the GeForce GTX 295 duo beats the GeForce GTX 285 trio, but the gap is a mere 2%. The scalability is all right, though. The performance grows almost 60% over the single card, which is quite good for a quad-GPU subsystem. However, it is the quad-GPU solution from AMD that wins the resolutions above 1280x1024 and enjoys a much larger lead over the Radeon HD 4890 despite having lower frequencies.
To achieve a playable speed in this game we disabled FSAA and such resource-consuming options as Sun rays, Wet surfaces and Volumetric Smoke. We use the Enhanced full dynamic lighting (DX10) mode for our test and additionally enable the DirectX 10.1 mode for the ATI cards.
It is the first time that the GeForce GTX 295 SLI tandem has no rivals. It is ahead of both the trio of GeForce GTX 285 and the duo of Radeon HD 4870 X2. The pursuers are close behind at 2560x1600, yet Nvidia’s quad-GPU monster doesn’t give up its lead. It offers a playable frame rate with some reserve for the most action-heavy scenes.
Today’s multi-GPU solutions are redundant for this game, yet the GeForce GTX 295 SLI is the winner again. It beats AMD’s quad-GPU as well as Nvidia’s triple-GPU platform.
The Radeon HD 4870 X2 4-way CrossFireX tandem wins this test, but the performance of Nvidia’s solutions is interesting, too. The transition from one GeForce GTX 295 to two such cards leads to a huge performance boost that amounts to 107% at 1920x1200. The bottom speed of this tandem is higher than that of the leading solution from AMD. It should be noted, however, that the GeForce GTX 285 3-way SLI is not much slower than the graphics subsystem with four G200b cores.
It is impossible to say what multi-GPU solution is better here at resolutions below 2560x1600. In the highest display mode AMD’s products deliver the best average speed, but Nvidia’s ones are not much slower. The difference is negligible when it comes to practical play. There is actually no reason to spend huge money for a graphics subsystem with three or four GPUs if this game runs just fine on a single dual-processor card.
Like in some other tests, the GeForce GTX 295 SLI cannot beat the GeForce GTX 285 trio, but the game is not as resource-consuming as to demand a multi-GPU solution to run quickly. The single GeForce GTX 285 ensures a generous reserve of speed at 2560x1600, so there is no point in upgrading your graphics subsystem any further to play Race Driver: GRID.
We use the in-game benchmarking tools that do not allow to measure the bottom frame rate. We also enable DirectX 10.1 support for ATI’s solutions.
This game doesn’t make a stable benchmarking tool, and its speed plummets on ATI’s GPUs when you turn on Adaptive AA, so the results should be viewed as approximations. Anyway, we can see that the GeForce GTX 285 trio is again faster than the GeForce GTX 295 tandem at every resolution, excepting 1280x1024.
The game has a frame rate limiter fixed at 30fps.
The GeForce GTX 285 beats AMD’s solutions at 2560x1600 but the GeForce GTX 295 SLI is unable to deliver a comparable bottom speed. On the other hand, the latter platform is superior to the other tested solutions in terms of bottom speed at 1920x1200.
It is at 2560x1600 that the GeForce GTX 295 SLI fails completely, being 13% slower than the GeForce GTX 285 2-way SLI in average performance and unable to maintain a comfortable bottom speed whereas the ordinary GeForce GTX 285 SLI copes with this task easily. This must be due to the lack of graphics memory because GeForce GTX 295 based solutions offer only 896MB of memory to 3D applications whereas GeForce GTX 285 based ones have 1024 megabytes.
This benchmarking suite is old and cannot reveal the full potential of the GeForce GTX 295 SLI at the default settings. The Radeon HD 4870 4-wat CrossFireX platform turns in the best overall score, but these results do not agree with what we have seen in real-life games.
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 display resolutions to provide a full picture.
The GeForce GTX 295 SLI tandem goes ahead and outperforms the GeForce GTX 285 trio – obviously thanks to the fourth GPU although the triple-GPU platform has often been superior to the quad-GPU one in real games. AMD/ATI’s solutions do not have software PhysX support and cannot compete with Nvidia’s products in this test.
In the second test the GeForce GTX 295 SLI enjoys a larger advantage over the GeForce GTX 285 3-way SLI than in the first one. Overall, the individual tests agree with the overall scores.
Alas, the quad-processor platform made out of two GeForce GTX 295 cards is a complete disappointment. And this is not the first time that Nvidia’s Quad SLI solution is such a failure. The two earlier generations of this technology, based on GeForce 7900/7950 GX2 and GeForce 9800 GX2 cards respectively, were not quite successful, either. Notwithstanding the fourth GPU, the Quad SLI platform was inferior to its triple-GPU counterpart based on GeForce GTX 285 in quite a lot of games.
Let’s view the results in more detail.
The average performance growth of the Quad SLI platform relative to the single GeForce GTX 295 is a mere 18.5%. It is only in three tests (S.T.A.L.K.E.R.: Clear Sky, Devil May Cry 4 and Prince of Persia) that the performance boost was higher than 60%. The Quad SLI tandem could only outperform the GeForce GTX 285 3-way SLI in five tests, its advantage being less than 15% in two of them. Nvidia’s solution is overall better than the Radeon HD 4870 4-way CrossFireX, but the latter enjoys a considerable advantage in three tests.
The resolution of 1680x1050 can hardly be interesting for top-end multi-GPU products but we can note that the GeForce GTX 285 3-way SLI looks better here than the GeForce GTX 295 SLI. The latter only has a large advantage in S.T.A.L.K.E.R.: Clear Sky. Otherwise, its advantage, if any, has no practical worth.
The resolution of 1920x1200 is quite popular among users of top-end gaming systems, but it is the GeForce GTX 285 3-way SLI that turns in the best results here. Assembling a graphics subsystem out of two GeForce GTX 295 makes no sense. You can even have a performance hit in some newer games, e.g. Call of Juarez: Bound in Blood, for which there are no driver optimizations as yet.
Well, it is the resolution of 2560x1600 that 3- and 4-way multi-GPU systems are meant for. Besides GPU performance, the amount of onboard graphics memory can be a factor here, and the GeForce GTX 295 SLI is inferior to the GeForce GTX 285 3-way SLI in terms of memory. The former subsystem is ahead by an average 2%, but is still worse in Call of Juarez: Bound in Blood. The difference between the two solutions cannot be felt in practice in S.T.A.L.K.E.R.: Clear Sky and Devil May Cry 4.
The GeForce GTX 295 SLI platform is good in terms of scalability. The doubling of GPUs leads to an average 47% performance growth, the peak performance boost being as high as 60% or even 90%. Building a Quad SLI system to play Crysis Warhead still has no point because no graphics solution available today can deliver a playable speed in this game at high resolutions. The level of comfort grows up, however, in such games as S.T.A.L.K.E.R.: Clear Sky and Prince of Persia, especially in action-heavy scenes.
Thus, the Quad SLI platform is formally the most advanced multi-GPU solution Nvidia offers today, but it is often inferior to the 3-way GeForce GTX 285 platform in practice. The latter is overall preferable, except when it is impossible to build (e.g. your mainboard has only two PCI Express x16 slots or the slots are located in such a way that you cannot install a third card into your system case).
Generally speaking, we consider Nvidia’s Quad SLI and 3-way SLI as well as similar solutions from AMD/ATI a compromise. They have a number of inherent drawbacks that can be hardly eliminated even theoretically. They are bulky, noisy, power-hungry and must be supported on the side of the driver and game. Buying a $1000 graphics subsystem, you want to have maximum performance out of the box, but today’s multi-GPU solutions can’t offer it. On the other hand, for some enthusiasts the very process of building, setting up and benchmarking a multi-GPU subsystem to set new 3DMark records is a pleasure, but that’s far from what gamers want.
Both graphics cards are exact copies of Nvidia reference design and differ from one another and the previously reviewed Inno3D GeForce GTX 295 Platinum only by the stickers on the cooler casing, package and bundled accessories. If you don’t really care about the accessories, then you can go for any of these three graphics cards – their all come with not too many goodies. Only Gigabyte GV-N295 comes with Molex → PCI express power adapters. Gigabyte solution overclocks a little better, but overclocking is in fact quite unpredictable and the same graphics card from the same manufacturer may demonstrate better or worse results in your particular case.
Highs:
Lows: