Drivers and practical details

Drivers

Activating SLI or CrossFire X (CrossFire X is now the name given to all CrossFire technology in Windows Vista) in drivers is very simple as you only have to check a box in their specific panel. Both AMD and Nvidia integrate a hardware and software version of their multi-GPU technology. The hardware version uses dedicated components present in GPUs as well as the specific connector to transfer information (or at least the largest possible part of it) from one card to another. This is the highest performance mode. The software version works with the PCI Express bus to transfer all the data between cards and is therefore a little less efficient.

For AMD, activation and deactivation is done very quickly. On the other hand, for Nvidia it takes longer. You must be patient and there are two confirmations not to be missed between turning off the screen and deactivating Aero.

Also for Nvidia, the driver has still not progressed in terms of multi monitor; in multi-GPU mode you can only use a single monitor while AMD has lifted this limitation. Another constraint is that if Quad SLI poses a problem in games while SLI otherwise functions well (and therefore a single 9800 GX2), you have to physically remove the card from the system to go into SLI. Deactivating Quad SLI also deactivates all multi-GPU modes and prevents you from taking advantage of the total power of a single 9800 GX2. In short, it’s all or nothing. For AMD, the Radeon HD 3870 X2’s GPUs function when quad CrossFire isn't activated.

Practical details

Of course, we used the hardware version in all cases, or at least placed the connectors to enable it to function. It’s not always easy to know what the driver is doing with more than 2 GPUs, especially for AMD, which moreover, wasn't too clear on the subject. On the one hand, they told us that the Radeon HD 3870 X2 uses exclusively the PCI Express switch for communication, but also that there is CrossFire cabling on the PCB. Obviously, at least one of these sources seems confused. So how about for triple and quad CrossFire? Either way, a connector is required for the driver to accept activation although we wonder if it is even used.

For CrossFire, a single connector is now required and it’s up to you to choose either one if your card has two. Otherwise, putting the two connectors into place doesn’t pose a problem. For Nvidia, a single connector is required for SLI or Quad SLI (which has always been the case), while a special one is needed for triple SLI which allows interconnecting all the cards (the 1st to the 2nd, the 2nd to the 3rd and the 3rd to the 1st). This is something that doesn’t come with nForce 680i motherboards and has to be acquired. NForce 780i and 790i models include it with their bundles.


CrossFire and 3-way SLI connectors.
In triple SLI, we may wonder why all the GPUs have to be connected the all the other ones while this isn’t the case in Quad SLI. The latter only uses one connector between the two pairs of GPUs instead of the four that would be needed for it to resemble triple SLI communication. This means that either one of the connections in triple SLI is useless (in this case, two classic SLI connectors would have been enough instead of a specific one), or triple and Quad SLI do not use the same communication architecture which means there is dead weight on the driver level.

We encountered several practical problems with SLI connectors when connecting the two GeForce 9800 GX2s. First of all, a flexible SLI connector, which barely covered the distance between the two cards (and therefore laying on top) partially melted and was damaged in hours of testing. The reason was that it was too close to the vent which evacuated hot air. We then exchanged it for a hard SLI connector and a new problem arose; contact was poor and caused the screen to blink. The casings of the 9800 GX2s are a bit higher than the PCB and consequently some of the relatively flat connectors (such as the reference versions found on nForce 680i and 780i motherboards) do not plug in correctly. The solution was to use a long flexible model so that it could make a loop on the top of the cards and not be in direct contact with the hot part of the casing.


The ‘’hard’’ SLI connector is flatter than the soft one and cannot be entirely plugged into the 9800 GX2 due to its casing

Specifications, consumption, the test

Specifications

Once again, note that memory quantities are not added up in multi-GPU systems. The 2 GB found in quad SLI or quad CrossFire are really only 512 MB in practice. This is a serious limitation in high resolution, which by the way, is the target of such systems.

Power consumption

We evaluated the power consumption of the different systems. Measurements were taken at the wall socket, in order to have the total power consumption of the power supply, in this case a Cooler Master Real Power M1000 (1000 watt). Note that motherboards were different for AMD and Nvidia solutions; an Asus P5E3 (X38) and Asus Striker II (nForce 790i). The latter proved to generally consume a little less (roughly 10 watts).


The use of the 55 nanometer process and PowerPlay to reduce consumption, enabled Radeon HD 3870 based solutions to be relatively economical in stand-by. We say “relatively” because with a four GPU configuration, there is still an additional 100 watts that is being wasted. In load, PowerPlay no longer gives the Radeons this advantage.
The GeForce 8800 Ultras still show record consumption mostly in stand-by where readings are almost 100 watts per card. In load, a couple of Radeon HD 3870 X2s or GeForce 9800 GX2s show similar results.

Given that values come close to 700 watts for a "basic" system, in other words, without overclocking and multiple hard drives and other fans, it’s obvious that there will have to be a serious power supply.

Note that the GeForce 9800 GX2, alone or in Quad SLI, is Hybrid Power compatible and can therefore be entirely turned off when used in a system compatible with this technology (such as the future nForce 780a for AMD processors). We will have to wait for summer until a similar alternative arrives for the Core 2.

The test

For this test, we used ten games, four of which support DirectX 10. Tests were carried out in 1920x1200 and 2560x1600, the resolution of 30" screens.
Anisotropic filtering as well as HDR were activated in all cases when available in the game. Finally, transparency/adaptive antialiasing were activated in multi-sampling mode.

Of course, all Windows Vista updates relative to performance and multi-GPU systems were installed.

Configuration

Intel Core 2GB Extreme QX9770
Asus Striker II nForce 790i (GeForce)
Asus P5E3 X38 (Radeon)
2 GB DDR3
Windows Vista
Forceware 174.53
Catalyst 8.3

Enemy Territory : Quake Wars

Enemy Territory : Quake Wars

While Quake Wars is based on the Doom 3 engine, it has undergone some evolution such as mega texturing which facilitates the work of artists; however, there is the additional cost in terms of decoding and access to mega textures. In the end, Quake Wars is a little more resource heavy than Doom 3 or Quake 4.
We saved a demo in a sequence versus 4 bots. Given that artificial intelligence was not calculated in the time demo, results were less affected by the CPU than in actual gameplay or at least in this case versus our bot adversaries.

All parameters were set to a maximum in the game including 16x anisotropic filtering. The patch 1.2 was used.


In this first test, SLI does better than CrossFire which with more than 2 graphic cards doesn’t add any gain. Triple and Quad SLI, however, are CPU limited especially in 1920x1200.


Once antialiasing is activated, the three GeForce 8800 Ultras surpass the two GeForce 9800 GX2s in 2560x1600. This is probably due to the fact that they are handicapped by their memory limited to 512 MB. Triple and quad CrossFire do not function correctly. It’s probable that the OpenGL driver is not yet "CrossFire X".

Half Life 2 Episode 2

Half Life 2 Episode 2

Still based on the Source Engine, Half Life 2 Episode 2 doesn’t really have anything new on the technological level. It simply optimizes and more heavily relies on the engine’s capabilities, making the game more resource heavy than its previous versions. We carry out a demo with all game options set to a maximum including anisotropic filtering which is in 16x.


In Half Life 2 Episode 2, CrossFire does better than SLI, especially in "low" resolution. Triple and Quad SLI have a problem here and performances are sometime reduced compared to classic SLI.


With antialiasing, CrossFire loses its advantage in 2560x1600 compared to GeForce 8800 Ultra triple SLI which also finishes ahead of quad SLI.

S.T.A.L.K.E.R.

S.T.A.L.K.E.R.

We carry out an identical movement and measure the framerate with fraps. The test was done in high quality, complete dynamic lighting, maximum details (anisotropic filtering 16x) and foliage shadows. S.T.A.L.K.E.R. uses an engine based on differed rendering, which is fundamentally incompatible with MSAA and makes the use of antialiasing impossible (at least this is what we thought). Either way, Nvidia ended up finding a way to apply it despite all of this! The 1.00005 patch is used.


CrossFire X shows interesting gains except for quad CrossFire in 2560x1600 which theoretically suffers from a lack of memory. Triple SLI is immune to this thanks to the 768 MB of the GeForce 8800 Ultras. Otherwise, the game systematically crashed when loading once Quad SLI was activated.


With FSAA 4x activated via Nvidia drivers, memory consumption increased while it was already very high to begin with. For this reason, cards equipped with only 512 MB (and therefore the 9800 GX2) are left behind in 1920x1200. In 2560x1600, the 768 MB of the GeForce 8800 Ultra are no longer enough. Finally, AMD does not offer anti-aliasing support in this game.

Rainbow Six : Vegas

Rainbow Six : Vegas

The first PC game based on the Unreal Engine 3.0, Rainbow Six : Vegas is still a very resource heavy game. We measure performances in the introductory scene. The HDR mode is activated as it is more or less obligatory as without it banding is very noticeable. Shadows are set to “low” because a higher quality in this domain lowers performance too much in certain areas.


Originally designed for the Xbox 360, this game seems to have a natural affinity for the Radeon HD which has a similar architecture to the game console’s graphic chip. It’s the same for CrossFire which shows consequent gains here, even if they are smaller with triple and quad GPUs. The different SLI configurations also show very good performances although even in 2560x1600 simple SLI is enough.


The game does not support anti-aliasing but Nvidia has implemented it to drivers as it has done with Oblivion. This is contrary to AMD who unfortunately didn’t make this effort. In 2560x1600, triple SLI adds additional fluidity which is quite appreciable. Finally, while triple and quad SLI are neck and neck in 2560x1600, the former is ahead in 1920x1200.

Oblivion

Oblivion

We saved a specific movement in order for it to be always identical and the test reproducible. Of course, HDR was activated.


All multi-card solutions are CPU limited here and a single graphic card is generally enough even if two add a little more comfort in 2560x1600.


Once FSAA 4x is activated, multi-card solutions become useful. Note the good results of the Radeon HD 3870 in CrossFire which is almost equal to the GeForce 8800 Ultra in SLI. Triple and quad GPUs are limited here by the CPU.

Colin McRae DIRT

Colin McRae DIRT

To test Colin McRae‘s latest opus which is very resource heavy we carried out a well defined sequence in high quality mode. Note that the activation of antialiasing is highly recommended given the way menus are rendered and because post processing effects amplify aliasing. The patch 1.2 was used.


CrossFire X does relatively well in this game and the two Radeon HD 3870 X2s are only surpassed by the three GeForce 8800 Ultras in 2560x1600. In 1920x1200, the added cost related to triple and quad SLI causes a slight performance reduction.


With anti-aliasing, 512 MB becomes insufficient in 2560x1600. While the Radeon HD 3870 X2 has respectable performances, they drop with more than two GPUs for both AMD and Nvidia. Of course, the GeForce 8800 Ultra benefits from its surplus of memory here.

Bioshock

Bioshock

The first game based on the Unreal Engine 3.0 to support DirectX 10, Bioshock has great graphics even in DirectX 9 mode while it is less resource heavy than Rainbow Six : Vegas. We carry out a well defined sequence of movement with all options pushed to a maximum.


The Unreal Engine 3.0 is definitely very appreciated by multi-GPU systems and gains are very high. Given the extremely high numbers, triple and quad GPU systems aren’t really of practical interest here.


Like with Rainbow Six : Vegas, Nvidia allows the activation of anti-aliasing for this game which doesn’t normally support it. For AMD, it’s only functional with a single GPU.


In DirectX 10 mode, GeForce 8 performances are similar to what we obtained in DirectX 9 mode. This isn't the case for CrossFire X which adds fewer gains while with a single GPU performances increase.


Anti-aliasing in DirectX 10 is only functional with Nvidia cards. Triple SLI is faster than quad SLI in these conditions.

Company of Heroes

Company of Heroes

Given that Company of Heroes received a DirectX 10 patch that adds a real plus on the graphics level, we decided to add it to our test protocol. All graphic settings were pushed to a maximum except for terrain details which remained on High (Ultra mode is reserved for DirectX 10). Textures were also limited to High, because the game indicates a lack of system memory for the GeForce in DirectX 9 mode with Ultra textures.

We run the integrated test on version 1.71.



Company of Heroes also shows very high gains in multi-GPU mode; however a single card is generally enough.


It’s the identical situation with antialiasing 4x activated except for demanding gamers that want to play in 2560x1600. Triple and quad SLI both finish first.


In DirectX 10, performances are clearly lower as the graphics engine is more advanced (moreover, the values on the x-axis were modified to scale). This time, two graphic cards are of use even in 1920x1200. In these conditions, the Radeon HD 3870 does not allow playing in 2560x1600 with satisfactory quality. Quad CrossFire only adds minimal gains and in DirectX 10 the game crashes when loading in quad SLI.


Triple SLI easily holds the lead.

World in Conflict

World in Conflict

Very resource heavy and with nice graphics, it’s only natural World in Conflict joins our test suite. We carry out the internal test with the patch 1.0002. All game options are pushed to a maximum. Catalyst 7.12 was used as the 7.11b version wasn’t correctly functioning in CrossFire.


CrossFire X shows interesting gains, but we have to point out that previous Radeons were very far behind GeForce cards in this game in DirectX 9 mode. While SLI does well, this isn't the case for triple and quad SLI which actually lower performances.


With anti-aliasing 4x, the situation is identical except for the GeForce 9800 GX2s which fall behind a little bit more than the others due to a lack of memory.


In DirectX 10, CrossFire produces strange results in 2560x1600. Triple SLI is limited by the CPU while quad SLI, for the 3rd time, causes the game to crash when loading the level.


With anti-aliasing, memory requirements are very significant and quad CrossFire pays the cost. Triple SLI does better.

Crysis

Crysis

An absolute must in terms of gaming, Crysis was tested with its 1.2 patch (optimized for multi-GPU systems). We carried out the internal test in DirectX 9 High mode (forced via the addition of ‘’–dx9’’ behind the executable) and in DirectX 10 Very High.


In 2560x1600, except for the GeForce 8800 Ultras in SLI, all cards suffer from a lack of memory which limits or reduces performances. CrossFire X adds gains with up to three GPUs but they are much less than with Nvidia solutions. Triple and quad SLI finish neck and neck.


With anti-aliasing 4x, don’t plan on playing in 2560x1600 as no solution offers this capability. Otherwise, in 1920x1200 CrossFire X has a lot of problems. Triple and quad SLI do better, as the first is more at ease due to its larger memory and despite the advantage in calculation power of quad SLI.


In DirectX 10 and with all details pushed to a maximum, no solution allows getting close to playing in 2560x1600. In 1920x1200, as long as you don’t concentrate too much on fluidity, it’s almost possible to play in triple or quad SLI.


Finally, in DirectX 10 and with anti-aliasing 4x, only the 3 GeForce 8800 Ultras manage in 1920x1200, while in 2560x1600 it was impossible to finish the test. Strangely, triple CrossFire X has more difficulty here than quad CrossFire X.

Recap of performances

Recap of performances

Although individual game results are interesting, especially when involving multi GPU systems, we calculated a performance index based on all tests with the same weight for each game. A score of 100 was given to the GeForce 9800 GX2 in 1920x1200.


On average, GeForce 8800 Ultra triple SLI is in the lead aided by the “0” obtained in S.T.A.L.K.E.R. by quad SLI. The four Radeon HD 3870 GPUs can’t rival triple SLI and are more equivalent to a couple of 8800 Ultras.


With 4x antialiasing filtering activated, it’s the same situation except the GeForce 9800 GX2s lose a little bit more ground in 2560x1600 due to their lack of memory. As for the Radeon HD 3870, their bugged ROPs in antialiasing have pushed AMD to use an alternative solution, which while it is very flexible, holds back performances. For this reason, they do not shine in these conditions.

Note that results obtained in S.T.A.L.K.E.R., Rainbow Six : Vegas and Bioshock were not added to final scores here because the Radeon HD 3870 is not capable of antialiasing in these games (except with a single card with Bioshock and in DirectX 9). You can consult a graph which otherwise takes these games into account here.


With DirectX 10 games, quad SLI isn't ideal because two out of four games had a problem with this technology. For this reason, its average score isn’t too encouraging and triple SLI dominates. Triple Radeon HD 3870 CrossFire finishes equivalent to a GeForce 9800 GX2 and therefore behind 8800 Ultra SLI.


In DirectX 10 and FSAA 4x, the GeForce 8800 Ultra easily dominates. In these extreme conditions, the 768 MB are very useful. Quad SLI also suffered here from crashes in two games and a lack of memory in another.

Note that the results obtained in Bioshock were not added to final scores here because the Radeon HD 3870 is not capable of antialiasing in this game. You can consult the graph which otherwise takes this game into account here.

Gains in 1920x1200

Gains en 1920x1200

We put the performance gains from tests into graph form so that they could be interpreted more clearly. They represent the gains of triple and quad GPUs when compared to dual GPU systems and enable us to see the interest of using more than two GPUs at the current time.

If there are no results, this means that the multi-GPU system had a bug that made the game unplayable. A gain of 0% indicates either that there is no gain or that it reduced performances. In the latter case, it would be deactivated by the user. Therefore there are no negative scores. Note that when comparing gains between Radeons and GeForces, you should keep in mind the difference in performance levels, meaning the former are generally less CPU limited.


Gains from Quad SLI aren’t too convincing and the 4th GPU doesn’t seem to add much compared to the 3rd.


S.T.A.L.K.E.R., Rainbow Six Vegas and Bioshock are not taken into account when calculating the averages because there is no antialiasing 4x support for them with the Radeons.


In DirectX 10, Crysis saves quad SLI’s good name.


You may recall that antialiasing doesn't work with Radeons in Bioshock.

Gains in 2560x1600

Gains in 2560x1600

We put the performance gains from tests into graph form so that they could be interpreted more clearly. They represent the gains of triple and quad GPUs when compared to dual GPU systems and enable us to see the interest of using more than two GPUs at the current time.

If there are no results, this means that the multi-GPU system had a bug that made the game unplayable. A gain of 0% indicates either that there is no gain or that it reduced performances. In the latter case, it would be deactivated by the user. Therefore there are no negative scores. Note that when comparing gains between Radeons and GeForces, you should keep in mind the difference in performance levels, meaning the former are generally less CPU limited.

2560x1600 should show the most significant gains:


Indeed, gains are higher here than they were in 1920x1200 which is first and foremost to the advantage of quad SLI.


S.T.A.L.K.E.R., Rainbow Six Vegas and Bioshock are not taken into account when calculating the averages because there is no antialiasing 4x support for them with the Radeons.


Quad SLI is penalized by crashes in World in Conflict and Company of Heroes.


You may recall that antialiasing doesn't work with Radeons in Bioshock.

CrossFire X : gains in 1920x1200

CrossFire X : gains in 1920x1200

We tested CrossFire X in more detail with different heterogeneous combinations in order to see if performances and the added flexibility was advantageous. We didn’t put these overall results into the "performance recap" graph because this would have made them incomprehensible. Here are the gains from the different solutions. On the one hand, there are those when compared to a Radeon HD 3870 and also when compared to a Radeon HD 3870 X2. In addition to combinations already found on the previous pages, we combined a Radeon HD 3850 256 MB with a Radeon HD 3870 as well as with a Radeon HD 3870 X2. This was also another occasion to look at the difference in gains between two Radeon HD 3870s and a Radeon HD 3870 X2 while keeping in mind that frequencies only differ slightly.

If there are no results, this means that the multi-GPU system had a bug that made the game unplayable. A gain of 0% indicates either that there is no gain or that it reduced performances. In the latter case, it would be deactivated by the user. Therefore there are no negative scores.

CrossFire X : gains in 2560x1600

CrossFire X : gains in 2560x1600

We tested CrossFire X in more detail with different heterogeneous combinations in order to see if performances and the added flexibility was advantageous. We didn’t put these overall results into "performance recap" graph because this would have made them incomprehensible. Instead, here are the gains from the different solutions. On the one hand, there are those when compared to a Radeon HD 3870 and also when compared to a Radeon HD 3870 X2. In addition to combinations already found on the previous pages, we combined a Radeon HD 3850 256 MB with a Radeon HD 3870 as well as with a Radeon HD 3870 X2. This was also another occasion to look at the difference in gains between two Radeon HD 3870s and a Radeon HD 3870 X2 while keeping in mind that frequencies only differ slightly.

If there are no results, this means that the multi-GPU system had a bug that made the game unplayable. A gain of 0% indicates either that there is no gain or that it reduced performances. In the latter case, it would be deactivated by the user. Therefore there are no negative scores.

CrossFire X : recap of gains

CrossFire X : recap of gains

After having compiled all the previous results, here are the average gains we obtained with CrossFire X in the various test configurations:


As we suspected combining a Radeon HD 3850, or more particularly its 256 MB, with a faster Radeon seriously limits performances. This is therefore a solution to avoid.

Also, we still noted differences in performances between a couple of Radeon HD 3870s and a Radeon HD 3870 X2. We believe that it comes from the fact that the X2’s GPUs only communicate via the PCI Express bus as there doesn’t seem to be a CrossFire connection on the PCB. Moreover, this is something AMD confirmed in the beginning; however, this manufacturer seems to have recently changed its mind telling us that there is indeed a CrossFire type connection that is present and functional. So what should we believe? While waiting for clarification from AMD, it seems that given our results the first answer is the most probable.

Conclusion

Conclusion

As we often say of multi-GPU solutions (and as tests go on we are convinced even more), they are reserved for the high end and hardcore users. This is all the more true for triple and quad GPU systems.

In our opinion, for less expensive cards, there is little interest in the update of these solutions when compared to the rapid evolution of products in the mid-range segment and the added cost put on platform level to offer this capability. As for the average Joe willing to buy the fastest system, coming with triple or quad GPUs, they will have to put up with certain inconveniences related to these solutions.


This leaves us with those users who aren’t afraid to ‘’get their hands dirty’’ in putting these systems into place, closely follow the evolution of drivers which may improve a lot overtime, especially thanks to this article, determine the level of interest of this technology in their favorite games. And of course they will disregard the uncertain behavior of such solutions in the future in order to simply enjoy owning such an extreme system.

More specifically, if such a configuration interests you, we recommend GeForce 8800 GTX / Ultra triple SLI. You may recall that its 768 MB of memory is precious help as the quantity of memory for each GPU isn’t cumulative. In very high resolutions, this poses a problem, which is somewhat ironic, because this is the intended use of such an ultra high end system. Quad SLI has not yet convinced us, partly for this reason, as well as the various problems we encountered in certain games.


As for CrossFire X based solutions, while their flexibility shows the overall robustness of AMD’s multi-GPU driver, it has to be said that only more or less homogenous solutions are of interest. We particularly appreciated the possibility of combining a mono-GPU card with a dual-GPU one, a capability which Nvidia does not offer. This flexibility can also be found on the software level with AMD’s enabling the use of multiple screens independent of the number of GPUs used at the same time. Nvidia allows only the use of a single screen.

While Nvidia has the highest performance system with GeForce 8800 Ultra triple SLI and AMD proves to have the most robust and complete multi-GPU system in its implementation, neither of the two have us totally convinced. There is still some work to do in order to get multi-GPU systems working smoothly!