GTX 590 and HD 6990: Tri and Quad GPU performance - BeHardware
>> Graphics cards

Written by Damien Triolet

Published on April 16, 2011


Page 1


The arrival of ultra high-end bi-GPU graphics cards from AMD and NVIDIA opens the door for quad GPU systems that are fairly easy to set up, provided, of course, you have the rather big wad of cash you’ll need to invest. We wanted to look at the gains you get from such systems as well as the tri GPU variants, both at high resolution and in surround gaming. Is a quad set-up worth looking at?

What platform for your quad system?
With the GeForce GTX 590 and Radeon HD 6990 you can set up a quad GPU system with just two graphics cards, which simplifies things a good deal in comparison to similar systems based on mono-GPU cards which require a specially designed motherboard.

Although any motherboard with two 16x PCI Express ports can support these systems, in practice it’s a good idea to keep a free slot between the two Radeon HD 6990s or GeForce GTX 590s. This space ensures that cool air can flow into the fan as, in contrast to the GeForce GTX 580s with the beveled back, their efficiency drops significantly when two cards are too close to each other. Ideally, we even advise you to go for a model with two slots between the graphics cards, ie a total of four slots between the PCI Express graphics ports as they’re in double slot format. This was what we had in our tests carried out on the Asus Rampage III Extreme.

You also have to make sure you have an appropriate power supply. With the GeForce GTX 590s in SLI, the Radeon HD 6990s in CrossFire X and the GeForce GTX 580s in 3-way SLI, you’ll need a 1250W model.

Here are the solutions we tested:

GeForce GTX 580
GeForce GTX 580 x2
GeForce GTX 580 x3
GeForce GTX 590
GeForce GTX 590 x2
Radeon HD 6970
Radeon HD 6990
Radeon HD 6990 + Radeon HD 6970
Radeon HD 6990 x2

Page 2
MSI GTX 590 MSI, Sapphire HD 6990

MSI supplied us with one of its N590GTXs and we also managed to get our hands on a Sapphire Radeon HD 6990.

The MSI N590GTX uses the reference GeForce GTX 590 design and only differs in terms of the sticker on the cooler card as well as the engravings on the two small plates at the back from which the NVIDIA logo has been removed:

MSI also respects the reference clocks and delivers its card with a CD for drivers, a manual, a rapid installation guide, a DVI to VGA adaptor, an HDMI to DVI adaptor, a mini-DP to DisplayPort adaptor and two 6-pin to 8-pin PCI Express power supply cable convertors. These adaptors will be particularly practical when setting up a quad SLI system as few power supplies, even powerful ones, have four 8-pin connectors.

Sapphire Radeon HD 6990
Like MSI, Sapphire makes do with the reference Radeon HD 6990 sticker:

Sapphire delivers its card with a CD for drivers, a manual, a DVI to VGA adaptor, a mini-DP to DisplayPort adaptor, a mini-DP to HDMI passive adaptor, a mini-DP to DVI Single Link passive adaptor, a mini-DP to DVI Single Link active adaptor and a CrossFire connector that is sufficiently long to support four slots. Note that Sapphire doesn’t supply an 8-pin power supply adaptor.

Sapphire respects the reference clocks and follows the AMD policy in terms of including the switch to allow access to the 450W bios. There’s a sticker on the switch and using it voids the guarantee.

Given this guarantee problem with the 450W bios, we used the 375W mode exclusively for this test. Note however that with some partners (XFX and Powercolor) activation of the 450W bios doesn’t void the guarantee.

Page 3
Specifications, the test


Combining just two Radeon HD 6990s gives processing power of over 10 Teraflops!

The GeForces are based on a more effective design but have fewer processing units and just half the processing power. The GeForce GTX 590s in SLI have only 5% more processing power than a 3-way GTX 580 SLI system. However they have 14% more memory bandwidth.

The test
For this test, we opted for 2560x1600, the highest single screen resolution, and 5760x1080 in surround gaming on three 1080p screens. Naturally we used maximum quality settings in all games.

We no longer show decimals in game performance results so as to make the graph more readable We nevertheless note these values and use them when calculating the index. If you’re observant you’ll notice that the size of the bars also reflects this.

The Radeons and the GeForces were tested with texture filtering at the “quality” setting. All the Radeons were tested with the Catalyst 11.4 preview 2 (8.84.2) driver. All the GeForces were tested with the 270.51 beta drivers.

Test configuration
Intel Core i7 980X (HT deactivated)
Asus Rampage III Extreme
6 GB Corsair DDR3 1333
Windows 7 64 bits
Forceware 270.51 beta
Catalyst 11.4 preview2 (8.84.2)

Page 4
Energy consumption

Energy consumption
Given that these are multi-card systems, we measured the energy consumption for the platform as a whole at the wall socket, both in idle and in the 3D Mark 06 pixel shader test, during which Prime95 was also launched at low priority so as to ensure an identical load across all cards:

At idle, the multi-GPU Radeon systems draw less power than their GeForce equivalents.

In load, the three highest end systems all exceed 1000 watts. Note that two GeForce GTX 590s draw the same amount of power as three GeForce GTX 580s.

Page 5
Starcraft 2

Starcraft 2

To test Starcraft 2, we launched a replay and measured performances following one player’s view.

All graphics settings were pushed to a maximum. The game doesn’t support antialiasing which is therefore activated in the control panels of the AMD and NVIDIA drivers. Patch 1.0.3 was installed.

In this first test, the CrossFire X isn’t very efficient. The gains are small in comparison to the Radeon HD 6970, giving a big advantage to the GeForces at antialiasing 4x. The quad also gives lower performance than the tri CrossFire X. The GeForces however give very low performance at antialiasing 8x, no doubt suffering from a deficit in memory.

To avoid any cheating in the game, the Starcraft 2 developers have barred the use of surround gaming.

Page 6
Mafia II

Mafia II

The Mafia II engine passes physics handling over to the NVIDIA PhysX libraries and takes advantage to offer high physics settings which can be partially accellerated by the GeForces.

To measure performances we used the built-in benchmarks and all graphics options were pushed to a maximum, except for PhysX effects accelerated by the GPU:

The GeForces have a big advantage here, benefitting from what appears to be a more significant CPU limitation on the Radeon side as well as the fact that the quad CrossFire X systems are inefficient.

Mafia II can detect that a large display area is made up of three monitors and therefore recognises the resolution as 3x 1920x1080 and not 5760x1080, which means it can automatically adapt the interface. For reasons unknown however, this detection system fails with a Radeon HD 6990 + Radeon HD 6970 pairing (triple CrossFire X) as well as with the GeForce GTX 590 pairing. The interface is then unusable and the rendering of the benchmark mode is only displayed on one screen.

On the GeForce side, activating antialiasing has no effect. We imagine that NVIDIA must have blocked the activation of antialiasing in its drivers to prevent the impact of lack of memory.

Page 7


Bulletstorm is one of the best in the current crop of games. Although only in DirectX 9 mode, the rendering is pretty nice, based on version 3.5 of Unreal Engine.

All the graphics options were pushed to a max (high) and we measured performance with Fraps.

The Radeons do particularly well in this game, especially with antialiasing 8x. The quad CrossFire X gives no advantage over the tri CrossFire X system however.

With antialiasing 8x, the GeForces clearly lack memory with the surround set-up.

Page 8
Crysis 2

Crysis 2

Crysis 2 uses a development of the Crysis Warhead engine optimised for efficiency. The game has few graphics options with just three main profiles: high, very high and extreme. We tested in extreme mode which includes antialiasing 4x as well as in the same extreme mode without antialiasing, via a modification of the command file.

We measured performance with Fraps on version 1.1 of the game. Note that three weeks after the release of Crysis 2, AMD still hasn’t produced a universal CrossFire X profile. After a long wait, AMD finally (and on a third party site) released a poorly optimised profile limited to the Steam version. A new version of the profile (11.3 CAP3) has now come out, again released very discreetly. This profile had no impact as far as we could see. Given that there are no gains with CrossFire X without an appropriate profile, we renamed the executable as fear.exe so as to force the use of the profile from another game. This is however a temporary solution which neither guarantees optimum levels of performance nor the absence of bugs.

The tri and quad SLI systems easily outdo the CrossFire X systems, taking advantage of the fact that NVIDIA has released the right profile.

Page 9
Crysis Warhead

Crysis Warhead

In spite of the release of Crysis 2, Crysis Warhead remains a reference in terms of graphics card performance. We tested it in its 1.1 hotfix version, at 64 bits and in ‘Enthusiast’ mode.

The Radeon HD 6900s are out front here, particularly at FSAA 8x, a mode in which they benefit fully from having 2 GB of memory while the addition of each GPU on the GeForce side actually slows performance.

2GB also seems insufficient at 5760x1080 AA8x.

Page 10
Far Cry 2

Far Cry 2

This version of Far Cry isn’t really a great development as Crytek made the first episode in any case. As the owner of the licence, Ubisoft handled its development, with Crytek working on Crysis. No easy thing to inherit the graphics revolution that accompanied Far Cry, but the Ubisoft teams have done pretty well, even if the graphics don’t go as far as those in Crysis. The game is also less resource heavy which is no bad thing. It has DirectX 10.1 support to improve the performance levels of compatible cards. We installed patch 1.02 and used ultra high quality graphics mode.

Although the GeForces are traditionally in front of the Radeons in Far Cry 2, they seem to suffer from a more significant CPU limitation with the more recent drivers.

In surround, the tables are turned and the Radeons, which are limited to 69 fps, trail a long way behind the GeForces.

Page 11


H.A.W.X. is a flying action game. It uses a graphics engine that supports DirectX 10.1 to optimise results. Among the graphics effects it supports, note the presence of ambient occlusion that’s pushed to a max along with all other options. We used the built-in benchmark and patch 1.2 was installed.

Multi-GPU systems bring significant gains in this game, though the fps is too low to make this any use in practice.

In surround, H.A.W.X. is limited to antialiasing 2x.

Page 12


The first game with DirectX 11, or more precisely Direct3D 11 support, we couldn’t not test BattleForge. An update added in September 2009 gave support for Microsoft’s new API.

Compute Shaders 5.0 are used by the developers to accellerate SSAO processing (ambient occlusion). Compared to standard implementation, via the pixel shaders, this technique allows more efficient use of the available processing power by saturating the texturing units less. BattleForge offers two SSAO levels: High and Very High. Only the second, called HDAO (High Definition AO), uses Compute Shaders 5.0.

We used the games built-in benchmark and installed the 1.2 build 304941 update.

The GeForces give very good performance here but are limited by their fillrate, which explains the almost zero cost when antialiasing is on. CrossFire X and SLI both give significant gains.

Strangely, without antialiasing the Radeon HD 6990s in quad CrossFire X do relatviely badly.

Page 13
Civilization V

Civilization V

Pretty successful visually, Civilization V uses DirectX 11 to improve quality and optimise performance in the rendering of terrains thanks to tessellation and in implementing a special compression of textures thanks to the compute shaders. This compression allows it to retain the scenes of all the leaders in the memory. This second usage of DirectX 11 doesn’t concern us here however as we used the benchmark included on a game card. We zoom in slightly so as to reduce the CPU limitation which has a strong impact in this game.

All settings were pushed to a max and we measured performance with shadows and reflections. Patch 1.2 was installed.

The GeForces do very well here while the tri and quad Crossfire X systems are ineffective.

Although Civilization V supports surround gaming, unfortunately its built-in benchmark tool doesn’t.

Page 14
S.T.A.L.K.E.R. Call of Pripyat

S.T.A.L.K.E.R. Call of Pripyat

This new S.T.A.L.K.E.R. suite is based on a new development of the graphics engine which moves up to version 1.06.02 and supports Direct3D 11 which is used both to improve performance and quality, with the option to have more detailed light and shade as well as tessellation support.

Maximum quality mode was used and we activated tessellation. The game doesn’t support antialiasing 8x. Our test scene is 50% outside and 50% inside and inside it’s surrounded with several characters.

For reasons unknown the game crashes systematically when loading when a quad CrossFire X system is used. NVIDIA has improved performance considerably with its most recent drivers, which allows the GeForces to take a big lead with multi-GPU systems.

The same goes for surround mode. Note however that antialiasing doesn’t have so much of an impact on the Radeons.

Page 15
F1 2010

F1 2010

The latest Codemaster title, F1 2010 uses the same engine as DiRT 2 and supports DirectX 11 via patch 1.1 that we installed.

We pushed all the graphics options to a max and we used the game’s own test tool on the Spa-Rancorchamps circuit with a single F1.

In F1 2010, the Radeons are particularly comfortable but the multi-GPU sytems are limited by the CPU.

There’s also a significant CPU limitation in surround. It’s difficult to say if the tri and quad GPU solutions are effective in this game.

Page 16
Metro 2033

Metro 2033
Probably the most demanding title right now, Metro 2033 forces all recent graphics cards to their knees. It supports GPU PhysX but only for the generation of particles during impacts, a rather discreet effect that we therefore didn’t activate during the tests. In DirectX 11 mode, performance is identical to DirectX 10 mode but with two additional options: tessellation for characters and a very advanced, very demanding depth of field feature.

We tested it in DirectX 11 mode, at a very high quality level and with tessellation activated. The modes tested were: no antialiasing, MSAA 4x and MSAA 4X + Depth of Field. Metro 2033 doesn’t support antialiasing 8x.

The Radeons benefit from having a 2 GB memory when Depth of Field is on.

In surround here, the Radeons have an advantage as of antialiasing 4x. Once Depth of Field has been activated, they too are short of memory.

Page 17
Performance recap

Performance recap
Although individual game results are obviously worth looking at when you want to gauge performance in a specific game, we have also calculated a performance index based on all tests with the same weight for each game. Given that we weren’t able to get it to run on the quad CrossFire X system, we discounted S.T.A.L.K.E.R. Call of Pripyat from our index. We also excluded Mafia II in surround mode as it doesn’t support this mode across all the solutions tested.

We attributed an index of 100 to the GeForce GTX 590 at 2560x1600 AA4x:

Hold the mouse over the graph to view the cards by performance at 2560x1600 AA4x.

Overall, the three GeForce GTX 580s carry the day, but only by a short head over the two GeForce GTX 590s. Although the quad CrossFire solution is down on the quad SLI solution overall, the Radeon HD 6990s benefit from having 2 GB of memory to take the lead with antialiasing 8x.

Hold the mouse over the graph to view cards by performance at 5760x1080 AA4x.

In surround gaming too, while the quad CrossFire X system doesn’t give much of a gain over the tri CrossFire X, the Radeons benefit from their 2 GB memory to take the lead once antialiasing 4x is on.

Page 18
Tri & Quad: scaling at 2560x1600

Tri & Quad: scaling at 2560x1600
We have grouped the gains obtained in the various games with the tri and quad GPU systems in comparison to the bi GPUs:

[ NoAA ]  [ AA 4x ]  [ AA 8x ]

While the NVIDIA and AMD tri GPU solutions are much of a muchness, when it comes to the quad GPUs, the GeForces give much better performance, though they do suffer from a lack of memory in some games when antialiasing 8x is enabled (or Depth of Field in the case of Metro 2033).

Page 19
Tri & Quad: scaling at 5760x1080

Tri & Quad: scaling at 5760x1080
We have grouped the gains obtained in the various games with the tri and quad GPU systems in comparison to the bi GPUs:

[ NoAA ]  [ AA 4x ]  [ AA 8x ]

In surround gaming, the quad SLI solution gives bigger gains than the quad CrossFire X. Unfortunately it’s penalised more heavily due to lack of memory.

Page 20

After running through these tests, it’s impossible to feel wholeheartedly convinced by the quad GPU solutions. Indeed the reality is a lot more complex than the few almost perfect scaling case studies that appear in the AMD and NVIDIA documentation show.

What with games which give very good results but aren’t much use in practice given the fact that the available bi-GPU solutions already provide more than sufficient performance, games which saturate the memory of the multi-GPU systems, games limited by the CPU and games which refuse to recognise the fourth GPU, such extreme graphics systems only come into their own with a few titles, such as Crysis, S.T.A.L.K.E.R. Call of Pripyat and Metro 2033 at 2560x1600 with antialiasing, as well as Battleforge and at least in part, Bulletstorm and F1 2010 in surround gaming. Of course, this is no longer the case if your screen is limited to full HD.

Although none of the solutions tested is completely convincing, there is one we can exclude right away: the quad CrossFire X based on two Radeon HD 6990s. In our view, this is the least effective solutions because of the fact that the drivers are not optimised to allow you to benefit from the fourth GPU. This solution also makes a great deal of noise, indeed we have never tested as noisy a system as this and remember this was with the “low energy consumption” bios… Unless you go for watercooling, we can only advise you to avoid this system.

Using a Radeon HD 6970 with a Radeon HD 6990 looks a better bet as things stand, especially as this combination costs a good deal less. Unfortunately it’ll still make a good deal of noise as the Radeon HD 6990 is far from quiet.

Among the NVIDIA solutions, the GeForce GTX 580 in 3-way SLI and the GeForce GTX 590s in SLI are comparable, with a small advantage going to the GeForce GTX 580s as the tri GPU yield is, as you’d expect, higher than the quad GPU. What makes the biggest difference here are the formats used. The GeForce GTX 580s take up 6 slots and require an appropriate motherboard but expel hot air out of the casing while the GeForce GTX 590s take up just 4 slots on any SLI compatible motherboard but only expel part of the hot air. A small advantage with the GTX 590s is that attributing one of the GPUs to GPU PhysX processing only monopolises 25% of the processing power compared to 33% for the GeForce GTX 580s.

While our preference would be for one of these NVIDIA solutions, they do suffer from the fact that their memory is limited to 1.5 GB. Although this is a relatively generous amount of memory, it’s not enough in extreme situations, which is to say in those very situations which have most need of such graphics power. None of these extreme graphics solutions really covers all the bases therefore. There are merely several imperfect systems to choose from depending on what you want to use it for and you’ll just have to accept that you can’t be sure how your choice will handle future games…

To end up with, we want to raise the issue of specific profiles for each video game, which are unfortunately still required. While there are few developers really interested in multi-GPU solutions, AMD and NVIDIA still haven't managed to implement robust generic solutions to support multi-GPU systems. How much of a gain multi-GPU systems bring therefore depends largely on how quickly the manufacturers get these profiles out. While NVIDIA has done pretty well, it has to be said that AMD still has an awful lot to do and could even do with overhauling its profiles policy entirely. Even if a developer hasn’t been particularly cooperative upstream, the fact that it took so long to deliver a fully functioning profile for a game such as Crysis 2 is totally unacceptable.

Copyright © 1997-2015 BeHardware. All rights reserved.