Crossfire vs SLI Performance Comparison Review

Bosco ccokeman - 2008-07-07 19:42:34 in Video Cards
Category: Video Cards
Reviewed by: Bosco   ccokeman   
Reviewed on: April 27, 2009
Sapphire, Nvidia

Introduction:

As single GPU video cards continue to get faster and deliver even more performance, are dual video card setups still a viable option for those looking for more performance? If one is good, two or more have to be better, right? In most cases two is better than one, and four is better than two; in most cases, not all. It would be nice to slap in another video card and see double the performance, but again that's not always the case. The cards will scale, and in some games you will see tremendous scaling, but in others you will wonder why you even bothered with the additional purchase. So what does it take to run a two, three, or even four GPU system? First off, you need to choose your video card and manufacturer, and then choose the motherboard that will allow you to run a multi-GPU setup. Sometimes that choice is made for you, if you just have to have a specific motherboard. But systems that were set up for Nvidia's multi-GPU solution SLI, or Scalable Link Interface, you needed a motherboard with an Nvidia chipset that supports the technology. For CrossFire based systems, you had to have a chipset that supports CrossFire technology. Simple enough! What if you already owned a motherboard that only featured a CrossFire capable chipset, and wanted to use Nvidia's multi-GPU solution? Well my friend, you were stuck buying a new motherboard. The solution was much the same for someone who wanted to run CrossFire and had a motherboard that only supported SLI. You get the drift. With the advent of the X58 chipset from Intel, you can have your cake and eat it too, since both technologies are supported.

There are two current technologies in use today. Nvidia's Scalable Link Interface, and ATI's CrossFireX technology. Both have their advantages, but the ultimate decision on which to buy and use is made by you, the consumer.

Both technologies offer improved performance, and will increase the cost of your system build. As part of that build, consider a monitor that is capable of running resolutions that can take advantage of the graphics horsepower delivered by these solutions. A 24 inch monitor capable of 1920x1200, or even a 30 inch monitor capable of resolutions up to 2560x1600, should be considered one of the minimum requirements when running a multi-GPU setup. If you are going to spend $1000 or more on video cards, another grand for a 30 inch monitor should be part of your build's budget. Enough chit-chat, let's see what kind of performance these kinds of setups can deliver. This comparison is not so much about the technology, but the performance you can expect for a system built with a multi-GPU solution.

 

Closer Look:

To run this sort of comparison, you have to have the video cards to do the testing. While there are only 11 different combinations being tested, it does not hit you until you sit down and unpack them all for a class photo as to how many there really are. A total of 16 cards all at once is a pretty big stack in my book. They range from the HD 4850 and GTX 260, all the way up to the GTX 295 and HD 4870x2. The only one left out was the GTX 275, as they just did not make it in time for the testing. So take a minute to bask in the ambiance of some high-end video horsepower. I had to sit back and just look for a while myself.

 

 

 

 

 

 

 

 

 

 

 

Enough looking, let's see what they can do! I want to see just how well the big boys perform without all the benchmark guru OS tweaking and extreme cooling, to see what the average guy or gal can come up with just by slapping a few cards into his or her pride and joy!

 

Testing:

Setup and Configuration:

The testing setup used in this comparison will be a Corei7 based system. Rather than run at the i7 965's stock 3.2Ghz, I will overclock it to almost 3.9GHz to try and eliminate a CPU bottleneck from being a concern. Each set of cards will be run with the latest drivers, ATI Cards using the Catalyst 9.3 driver - and the 9.4 version used only for the HD 4890 - as all of the ATI testing was completed before the HD 4890 was released. On the Nvidia side, the driver used is the latest 185 series driver, which fixes the Dead Space issue with the Gateway XHD 3000 monitor. Each respective control panel will be left at factory default settings, with the exception being that PhysX will be turned off in the 3DMark Vantage test, since the Orb will not publish any scores with PhysX enabled. PhysX will be enabled in the Nvidia Control panel and in games at all other times. There will be 3 separate classes in this testing. The dual GPU cards in a quad GPU setup, the same cards individually, and the single GPU cards in a multi-GPU setup.

Testing Setup

Testing Setup i7:

 

Comparison Video Cards:

 

Testing:

Far Cry 2:

Featuring a new game engine named Dunia, this game looks to be another one to stress your video card. Built especially for Far Cry 2, this engine allows for real time effects and damage. This next generation first person shooter comes to us from Ubisoft surprisingly - not from Crytek. The game is set in a war-torn region of Africa where there is a non-existent central government and the chaos that surrounds this type of social environment. If you have seen the movie Blood Diamond, you know the setting. Ubisoft puts the main storyline of the game into focus with these statements: "Caught between two rival factions in war-torn Africa, you are sent to take out "The Jackal," a mysterious character who has rekindled the conflict between the warlords, jeopardizing thousands of lives. In order to fulfill your mission you will have to play the factions against each other, identify and exploit their weaknesses, and neutralize their superior numbers and firepower with surprise, subversion, cunning and, of course, brute force." In this version of the game, you don't have the beautiful water, but instead the beauty and harshness of the African continent to contend with. Most games give you a set area that can be played through, while Ubisoft has given the gamer the equivalent of 50km2 of the vast African continent to explore while in pursuit of your goals. The settings used are just a few steps below the maximum in-game settings and offer a good blend of performance vs. visual quality.

 

Settings:

 

 

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

The Far Cry 2 testing shows that at 1920x1200, the HD 4870x2 CrossFireX combination proved to be a bit stout for everything but the GTX 285 SLI setup. The tables turned on the ATI camp at 2560x1600, with the HD 4870x2 dropping 29 frames per second to 71, 5 fewer than the GTX 295 Quad SLI setup that lost only 14 frames per second with the resolution change. What is surprising is that the GTX 285 combo kept up with the Quad SLI setup at the top end.

 

Testing:

Crysis, released in late 2007, has been around long enough for graphics card technology to evolve to a point where playable frame rates can be achieved without having to resort to a low detail environment. The game was developed by Crytek and is published by EA Games. This Sci-Fi First Person Shooter has you playing as part of an elite team sent to respond to a distress signal on a small tropical island in the Philippines.

 

 

 

 

 

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

While still a stout test when the settings are increased, the quad GPU setups give frame rates that are entirely playable, from just under 60 FPS at 1920x200 to just over 50 FPS at 2560x1600. The single GPU cards were able to deliver decent performance even at 2560x1600, with the exception being the CrossFire'd HD 4850s.

 

Testing:

Crysis Warhead is a standalone expansion pack situated in time with the storyline of the original Crysis. As Sergeant "Psycho" Sykes, you have a secret mission to accomplish on the far side of the island. Along the way, there are EMP blasts and aliens to contend with, as you hunt down the KPA chief. This game uses an enhanced version of the CryEngine 2.

 

 

 

 

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

With AA at x2 and the balance of the settings at the "Enthusuiast " level, this game is still a stiff test for the cards used in this roundup. The GTX 295 makes the best of this at 1920x1200, and drops 30 FPS when the resolution is increased to 2560x1600 - where the performance differential with it and the CrossFireX HD 4870x2 combo is really reduced to 2 FPS. In this test, the GTX 285 SLI setup fared better at 2560x1600 than the single GTX 295.

 

Testing:

PT Boats: Knights of the Sea is a DX10 title that features its own proprietary graphics engine currently in development. The game is a combination of Real Time Strategy and Simulation. You have the ability to control the entire crew or just a single member. Play as the German, Russian or Allied navies, and prove your mettle on the open seas.

 

Video Settings:

 

 

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

When the AA is turned on, this game usually tanks the ATI GPUs - resulting in really low FPS, as seen with the HD 4850 CorssFire combo; the rest of the ATI combos seem to hold out much better. With the upper end combos, the GTX 295 wins in both the quad and dual GPU modes in both resolutions. The GTX 285 blows away the HD 4890 and HD 4870 combinations again at both resolutions. The scoring gets a little closer when the AA is turned off, but that's not what this test is about.

 

Testing:

BioShock is one of the creepier games you can play. The building of a perfect Utopian society undersea gone horribly wrong. Its inhabitants driven mad with the introduction of tonics and genetic modifications. Now Rapture is just a shadow of its former glory, with little girls looting the dead of what little they have left while being shadowed by guardians known as "Big Daddies". It is a demanding game that will make your hardware scream for mercy. This First Person Shooter allows for an infinite number of weapons and modifications to provide a unique experience each time it is played. The environment, as well as the story line, will wrap you up for hours on end.

 

Video Settings:

 

  

 

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

At 1920x1200, the HD 4850x2 CrossFireX setup delivers an identical average FPS as the HD 4870x2 CrossFireX combo; it does trail off once you move to 2560x1600. Both manufacturers' video card systems scale well in this game. The GTX 295 shows a 26 FPS improvement at 1920x1200 that blossoms to almost 100FPS at 2560x1600. The scaling by the HD 4870x2 is a little better at 1920x1200, with an increase of 67 FPS, and a 74 FPS increase at the maximum resolution.

 

Testing:

World in Conflict was released in late 2007, and is a Real Time Strategy game. The difference in this RTS game is that it is not the typical "generate wealth and build" type of game. Instead, you advance by conquering your foe.

 

Video Settings:

 

 

 

 

 

 

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

CrossFire scaling in World in Conflict is non-existent, and even shows negative scaling at 2560x1600 on the HD 4870x2, while the 4850x2 does show scaling. In all three categories, and both resolutions, the FPS delivered was highest with the Nvidia cards. At 2560x1600, the HD 4850x2 and HD 4850 CrossFire setup fall right into what would be considered by most people to be unplayable frame rates. The HD 4890 CrossFire and GTX 260 SLI combos match up well in World in Conflict, with no real difference in the performance.

 

Testing:

Cryostasis: The Sleep of Reason is a survival horror First Person Shooter that uses Nvidia's PhysX technology to simulate realistic movements of cloth, fluids, fog, smoke and destructible environments. The fluid simulations are the highlight of this game.

Settings

 

 

 

 

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

This benchmark is quite demanding when PhysX calculations are enabled. The ATI cards hover in the mid teens for FPS delivered across all resolutions and all card combinations, while the Nvidia cards did show scaling and delivered FPS based on their capabilities.

Testing:

IL-2 Sturmovik: 1946 is a World War II Flight Simulator situated on the eastern front. The game has seen steady development, and with its sequels boasts a large community of gamers. Expansions include Forgotten Battles, Pacific Fighters and 1946. This version includes new maps and an increase in the amount of planes that can be flown. The game was originally released in 2001 by Ubisoft and was developed by 1C Maddox Games. With the current patches and custom configuration files, the game still provides a stern test at higher resolutions on single and dual video card setups. The game features pre-flown missions that can be used as a test for hardware, because the test is repeatable and does not vary, as in-game play would not allow this. When flown in real time, the performance mirrors the demo results.

Settings

 

 

 

Higher Scores = Better Performance

 

There was absolutely no CrossFire scaling from the HD 4870x2 or HD 4850x2 when a second card was added to the equation. On the other hand, the GTX 295 did show an increase in FPS with the addition of a second card; not much, but an increase always helps. The GTX 260 SLI combo did exceptionally well in this test, eclipsing the performance of the GTX 285 setup at 1920x1200. While this is an older game, it is still a popular Flight Sim with a large online community.

 

Testing:

Call of Duty 4: Modern Warfare is the successor to the Call of Duty crown. This iteration of the game is fought in many of the world's hot spots with modern armaments and firepower. You can play as either a US Marine or British SAS trooper. Since this game does not feature an in-game test, I will run through a section of the game and measure average FPS using Fraps 2.9.3.

 

Video Settings:

 

 

 

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

The quad GPU setups are just massive overkill if Call of Duty 4 is your game of choice at both 1920x1200 and 2560x1600. 300 plus FPS is just sick at 1920x1200 with the GTX 295 Quad SLI setup. At no point did the ATI cards come close to the level of performance delivered by the Nvidia cards across both resolutions and all three classes. Not one combo fell below 110 FPS.

 

Activision's Call of Duty: World at War goes right back to the bread and butter of the franchise - WWII FPS action. In this rendition, you start off in the South Pacific and move through a series of missions that flip back and forth between the Russian front and an island hopping advance toward the Imperial Japanese homeland. Included is a mission on Peliliu Island, arguably one of the more difficult and costly battles in the Pacific theater. The gameplay in the single player mode is rather short, but the game makes up for this shortcoming in online gameplay. If you thought COD4 looked nice, this game is amazing with the graphics maxed out playing at a high resolution. This game just may be my reason to move to a 30 inch monitor. I will use Fraps to measure a section of gameplay in the Semper Fi map on Makin Island to compare performance of these video cards.

Settings:

 

 

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

In most cases, the scaling is not as dramatic as it is here with the ATI dual GPU cards. At 2560x1600, the scaling is almost 100% with both the HD 4870x2 and the HD 4850x2. Overall, the highest FPS was delivered by the Nvidia contingent in all three categories at both resolutions. The 27 FPS differential between the GTX 295 Quad SLI setup and the HD 4870x2 CrossFireX dropped to only a 10 FPS advantage at 2560x1600.

Testing:

Warhammer 40,000: Dawn of War II is a Real Time Strategy game that is significantly different than its predecessor, with improved AI and an improved physics engine. You can play either as a single player in campaign mode, or in a multiplayer game where Microsoft's Live ranking system can be used.

Settings

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

In Dawn of War II, the quad GPU champ is the GTX 295 combo at 1920x1200 and 2560x1600. The GTX 285, however, is beaten soundly by the HD 4890 CrossFire setup at 1920x1200, while the performance draws even at 2560x1600. There is not a single setup here that results in a game that is unplayable.

Testing:

F.E.A.R. 2: Project Origin is the sequel to F.E.A.R. and was just recently released by Warner Brothers with development work by Monolith. This sequel starts off with Becket (the player you control) on a mission that takes place just before the finish of the original game. The first mission is to rescue Genevieve Aristide. F.E.A.R. 2 includes many of the elements made popular by the first iteration of the game with improved AI.

Settings

 

 

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

At 1920x1200, the ATI cards seem to have a distinct performance advantage over the Nvidia cards. At 2560x1600, the advantage has been nullified, and the Green team performs equally or slightly better in all three classes.

Testing:

Lost Planet: Extreme Conditions Colonies Edition is a Third Person Shooter that takes place on the fictional world E.D.N.III after Earth has become too inhospitable for the Human race. The game was released in the USA in early 2007, and was developed and distributed by CAPCOM.

 

Settings

 

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

The latest Catalyst drivers helped the ATI cards, but they still did not post the results that the Nvidia cards did. The GTX 295-based Quad SLI setup easily outperformed the HD 4870x2-based quad GPU setup in both resolutions. When it came to the single GPU cards, the GTX 285 is the top performer, with the HD 4890s falling down to the level of performance of the GTX 260 in both resolutions. The 260 is slightly slower than the HD 4890, and slightly faster than the HD 4870 1GB - not that the difference is enough to notice in-game.

Testing:

Mirror's Edge was released to the PC community on January 13, 2009. Mirror's Edge is designed using Unreal Engine 3. One of the highlights from EA and Dice was the fact that this is the first game to feature GPU accelerated PhysX effects throughout the game. As you may or may not know, PhsyX is becoming a huge part of the gaming experience. The additional realism that comes from using this technology on hardware designed to maximize the potential of PhysX capabilities makes playing the gaming that much more intense. From glass that stays part of the environment to the cloth that tears and moves realistically to the physically simulated smoke, this First Person Runner, as it is called, has plenty of hype regarding the stunning visual effects.

 

 

 

 

 

 

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

While enabling PhysX may seem unfair, this is about what combination of GPU cards perform best with the eye candy on. In Mirror's Edge, this results in the CPU doing the PhysX work with the ATI cards, allowing the Nvidia cards to post higher FPS averages, as a result of the GPU doing the PhysX calculations. The ATI cards all ran in the low 20 FPS range, with the exception of the HD 4850 CrossFire combo. The Green team's performance was significantly higher. I could have disabled PhysX, but then the effects are not as evident, and the systems do not struggle.

Testing:

Fallout 3 takes place after the nuclear holocaust that nearly wipes out civilization and leaves the world an irradiated mess. The vault, or fallout shelter, you are born in is Vault 101, situated in the Washington DC, Maryland and Virginia area. The premise of the game is that the Vault has been sealed for 200 years and now your father has opened the vault and escaped without a trace. The Overseer believes you are involved, so you must escape as well into the wasteland that was once our nation's capital. I find myself looking for landmarks since I am familiar with the streets of Washington DC.

Settings:

 

 

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

At 1920x1200, there are only 7 FPS separating the top and bottom cards in this comparison. Not really much to go on here, but the Nvidia combinations appeared to deliver higher FPS on average than the ATI offerings, with all of the Green setups in the 90 plus FPS range, and only the HD 4890 combo breaking into this range at 1920x1200 for the Red team.

 

Testing:

In Dead Space, as part of the crew of the USG Kellion, you are headed on a repair mission to repair a ship in distress. Things go from bad to worse, starting with the crash landing and seemingly silent and "Dead" ship, the USG Ishimuru. Offering a non-traditional over the shoulder viewing angle, the game gets right into the action as soon as the ventilation systems are activated. From there, things get worse with the appearance of the Necromorphs. Survival now becomes a primary concern for the primary character, Isaac Clarke. Survive, and you may find the loved one that was aboard the Ishimuru.

Settings:

 

 

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

In Dead Space, there is no doubt which multi GPU system the game is optimized for. The game is playable at 2560x1600 by all 11 combinations, with the in-game settings set as high as they go. The Nvidia cards just perform at a level that is much higher than that of the ATI cards.

 

Testing:

Left 4 Dead is a new release from Valve that leaves you as part of a group of survivors in a world where an infection has rapidly turned the populace into a zombie horde. You goal is to make it to a rescue point, all the while fighting what seems like overwhelming odds. Along the way there are safe houses where you can replenish your weapons and health. The movie 'I Am Legend' comes to mind to set the stage for this game. But unlike the movie, there are four characters and not just a lone gun and his faithful companion. The horde is not at all like the typical slow walking, foot shuffling zombie. These zombies are quick and work with the pack mentality. Your job: survival! .

Settings:

 

 

 

 

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

In the quad GPU testing, the difference between Red and Green cards is 6 FPS at 1920x1200, and 16 FPS at 2560x1600. At 1920x1200, the 4850x2 CrossFireX setup performs within 10 FPS of the HD 4870x2's. The GTX 285 outperforms the HD 4870x2's in both resolutions, and outperforms all of the standard multi GPU combinations in both resolutions.

 

Testing:

S.T.A.L.K.E.R.: Clear Sky is the prequel to the immensely popular S.T.A.L.K.E.R.: Shadow of Chernobyl, and is developed by GSC Gameworld. This First Person Shooter features an updated X-Ray graphics engine that includes DirectX 10 support and improved AI. You play Scar, a S.T.A.L.K.E.R. that patrols the "Zone".

 

Settings

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

The top performing multi GPU combo in this benchmark is the GTX 285 at both resolutions. The quad GPU giants just don't perform well enough to overcome the GTX 285. With that said, the quad GPU setups offer very little in the way of a differential in performance from one manufacturer to the other in both of the resolutions tested. The difference in performance from ATI to Nvidia is one to two frames per second. Both SLI and CrossFire seem to scale in this game, offering a performance increase, and making a dual GPU setup worthwhile.

 

Testing:

3DMark06 is one of the benchmarks that always comes up when a bragging contest is begun. 3DMark06 presents a severe test for many of today's hardware components. Let's see how this setup fares. The settings we will use are listed below.

 

Settings:

 

 

 

 

 

 

  

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

The ATI cards have a strong showing in 3DMark06, with the HD 4870x2 and 4850x2 CrossFireX combinations, beating out the GTX 295 Quad SLI setup, as well as the dual GPU card testing at the 1920x1200 level. The GTX 285 and GTX 260 dual GPU setups are beaten at the 1920x1200 level as well. At 2560x1600, the performance differential shrinks between the GTX 295 and HD 4870x2 single card setups. The GTX 295 is the highest scoring in the standard multi-GPU class, outperforming the HD 4890 and HD 4870 1GB overclocked card from Sapphire at the 2560x1600 level.

Featuring all-new game tests, this benchmark is for use with Vista-based systems. "There are two all-new CPU tests that have been designed around a new 'Physics and Artificial Intelligence-related computation.' CPU test two offers support for physics related hardware." There are four preset levels that correspond to specific resolutions. 'Entry' is 1024x768, progressing to 'Extreme' at 1920x1200. Of course, each preset can be modified to arrange any number of user designed testing. For our testing, I will use the High and Extreme  presets at all default settings.

 Settings:

 

 

 

 

 

 

 

 

 

 

 

 

 

Higher Scores = Better Performance

 

The GTX 295 Quad SLI combo is the highest performer in the dual GPU card class, at both the 1920x1200 and 2560x1600 resolutions. The story is the same with the dual GPU cards run in single card mode. The next class down contains the single GPU cards running in multi-GPU mode - Crossfire or SLI. The GTX 285 and the GTX 260-216 combos are at the top of this field, with the Sapphire Toxic HD 4870 1GB card in a relatively close third position. The Nvidia cards are easily the top performers in this benchmark, much the same way the ATI cards are in the 3DMark06 testing.

Testing:

Power consumption is something that is starting to creep into everyone's decision making process when making a purchase. Of course the driving force here is just how much are you going to pay the electric company at the end of the month? Increased heat output from dual cards and the increased power consumption with dual cards are just two things that we are now thinking about when it comes down to the moment we spend our hard earned money on that high-end video card. Idle testing was not done, as the largest power draw is going to be when the cards are under a load. The test setup used is a system fairly typical of that seen in the enthusiast class, with a well overclocked, water cooled CPU, one HDD, an optical drive and not a whole lot else. The system idles right around 200 watts with a low-end video card installed as a baseline measurement. Testing will be done using 3DMark06 to place a load on the GPU's. I will loop "Canyon Flight" four times, noting the highest power consumption through each run, and average the four runs to reach the final power consumption numbers.

 

 

 

 

 

 

 

 

If you do any cruising through the video card threads on any number of web sites, the controversy is always price versus performance. Some people will always pay for higher performing cards, while you have the rest of the world who look for the balance and try to get the most bang for their buck. People buying the top-end cards with their top-end prices is always going to happen. But, when it comes to multi-GPU systems, the thought is usually on performance and not so much on cost. I remember my first multi-GPU system was with two ATI 1900XT's; the total cost was almost $1200 when the cards first came out. As an early adopter, I paid a premium for these cards - but at the time, this combination was pretty much top of the heap, until the Nvidia G80 cards came out and just cleaned house. Thinking about that $1200+ price tag and the level of performance delivered, the highest price combo in this comparison comes in at just over 1100 bucks, more than $100 less than my 1900XT combo from just a few years back. Just thinking of how far performance has come, and the costs associated with that performance, over the last few years is astounding.

 

Cost will always be a concern, so along that line I decided to see what the cost per FPS would be for each combo. To do this, I will add up the total FPS per card at each resolution and divide that total by 18, the amount of tests run where a result was given in FPS. The cost for the combo will then be divided by the average FPS to give a cost per FPS - effectively giving us a price versus performance comparison. Pricing will be the cards' current cost from Newegg. The measurement used will be dollars and cents.

 

 

The most power consumed by any combination in this test is hands down the HD 4870x2 CrossFireX combo, weighing in at 936 watts. This is almost 200 watts higher than the GTX295 in Quad SLI. The 4850x2, of course, had the lowest power consumed in the quad GPU category. When the x2 cards were run in single card mode, power consumption dropped as expected, with the GTX 295 still outperforming the HD 4870x2 - and almost using less power than the 4850x2. In the two GPU category, the GTX 285 SLI combo used the most power - but also delivered the highest performance. When it comes to pricing, the least expensive quad GPU setup is the HD 4850x2 CrossFireX combination, coming in at 520 bones. The GTX 295 is far and away the most expensive setup, but does prove its worth throughout the testing. Usually you get what you pay for. The least expensive dual GPU setup in this comparison is the Toxic HD 4850 setup at $260, and performance-wise the adage holds true - as this combo delivered the lowest performance of the group. When it came time to figure out which combo offered the most bang for your dollar, the GTX 295 Quad SLI setup is just about out of the running at $9.10 per FPS at 2560x1600. Even with the exceptional performance, the price tag is a bit steep at $1120 for the pair of cards. The HD 4870x2 setup is $320 less expensive, and begins to compete with the other combinations. The hands down best bang for your buck based on the testing I have done here is the GTX 260-216 SLI setup, which costs less than any other combination based on total cost versus performance delivered.

 

Conclusion:

ccokeman's Thoughts

So what did we learn from this little exercise? I learned that SLI is the more mature multi-GPU solution currently. In each of the three classes, the Nvidia technology and their video cards lost no more than four times out of 40 tests, with all things being equal - this being in the quad GPU class. Each of the other classes delivered a 37 to 3 margin of victory for team Green. Each set of cards was run at the default factory settings in the control panel to show what Joe Average will get out of a plug and play system. Could performance be improved upon by manipulating the settings in the control panel? Sure it could - on both counts, but this was about running what the system defaults to. The only deviation from this is in 3DMark Vantage, as the Nvidia cards share a distinct advantage in the PhsyX tests. The reasoning is that Futuremark does not allow the scores when the GPU does the PhysX calculations, because ATI does not have an equivalent technology to compete right now. Right now, SLI is the way to go for multi-GPU performance in the games tested in this comparison. The downside to this is that it will cost you a little bit more to get this performance, as ATI seems to have the pricing game down pat at the performance levels I looked at here. With just raw cost as a factor the GTX 295 quad GPU setup will set you back just over $1100, while the HD 4870x2 combo will only (insert sarcasm here) set you back $800-$1000 depending on the cards you buy. The Sapphire cards will only set you back 400 bucks each, instead of the more popular pricing of 500 bucks. In the single GPU card class, the GTX 285 combo will cost you almost $700 - at a cozy $680 - while the HD 4890 combo only goes for $530, with prices scaling downward based on the video card's capabilities. That being said, the best value per frame per second delivered is the GTX 260 SLI combination at $3.44 per FPS at 1920x1200, and $4.59 per frame at 2560x1600. In most of the games tested, the GTX 260 SLI combo did not deliver the best performance, but it was able to hold its own. Raw performance goes to the Quad SLI GTX 295 combo.

For most of us, at some point the power bill gets to be a concern - especially when you run a distributed computing project on both your CPU and GPUs. The power company wants their piece of the pie just like everyone else. While our test systems are not the most power hungry, it pulls a decent amount of power under load. The loser in the power consumption testing has to be the HD 4870x2 CrossFireX setup. With our systems and these two cards under load, the system pulled a total of 936 watts from the outlet. As a comparison, the Quad SLI setup used 757 watts from the mains. This scenario continued when the dual GPU cards were pulled out of multi-GPU mode, with the 4870x2 pulling 656 watts, and the GTX 295 pulling 471 watts under load. In the third class, the GTX 285 SLI combo did pull more current than the HD 4890 combo, with the GTX 260 setup falling between the HD 4870 and HD 4850 CrossFire setups.

When you look at the scaling in performance you get from going to a quad GPU setup, it is not going to be anywhere near 100% in most cases. There were a few instances where the scaling on the HD 4850x2 and HD 4870x2 scaled close to 100% - which was a real surprise. This was the exception, and not the rule. In turn, there were games where there was no scaling whatsoever with cards in CrossFireX, while the SLI combo scaled well. That's something that could be a driver fix, to enable better CrossFire support in games. To use all this video horsepower, you will need a monitor that runs no less than a 1920x1200 resolution, as anything else means you are just throwing money away. The ideal solution is to go with that 30+ inch monitor you have been lusting over to take advantage of the firepower. Currently, Nvidia and its SLI technology is the performance winner here, on this system, with the drivers and video cards tested. The numbers are the numbers, and this is what they show. ATI has great price points, but they just can't deliver overpowering performance for that price. On the other hand, Nvidia delivers the performance but at a steeper price. The old adage "you have to pay to play" comes to mind here. By paying, you have a GPU that is capable of doing much more than just creating pretty pictures for us to look at on that magic screen. With Nvidia's CUDA technology, you have a wide array of applications ready to take advantage of the parallel computing capabilities of its architecture, such as vReveal from Motion DSP which allows you to clean up poorly shot video, Badaboom for converting video to most popular mobile formats, Folding@Home where performance is just amazing, with more coming each week, it seems. You have GeForce 3D Vision to immerse you in the game. GPU Acceleration in Photoshop! What more do you need? Drivers? Why yes indeed, Johnny. Nvidia seems to be more committed to delivering drivers almost as fast as I change my underwear (yes, it's a daily occurrence), while ATI is still stuck on a once a month schedule, and you need to hope and pray they work. One thing that could overcome the performance problems is having user adjustable CrossFire profiles so the performance is there. When there is no scaling or negative scaling, a profile for that game may be all that's needed to excel, but currently that's not the case. Price is a point of difference, but there is more to it than a GPU that costs less but offers less. You don't pay Hyundai dollars and expect Corvette performance. Sometimes you get a surprise, but now its ATI's turn to swing for the fences. The potential is there on the CrossFire side of the fence.

 

Bosco's Thoughts

To tap into what Frank said, what did I learn? That's an easy question to answer: consistency!

Nvidia has been very aggressive with getting their drivers to perform, either before a game is launched or on the launch date of the game. They have strived to be very aggressive, and it shows in our testing. When we all talk about multi-GPU performance, opinions vary on which company is better, and with ATI's driver issues it's kind of hard not to get frustrated. I have been very vocal about this in the past, and during our testing I was burning the phone lines up with Frank again complaining.

As Frank said, pricing is a big thing here, especially with it comes to running GTX 295s in Quad SLI, but really, if you are going to spend over $1000 on a 30" LCD, what's another $1000+ on video cards...seriously. If the situation were reversed, nobody could sit here and tell me that ATI would not charge $1100 for their cards. The reason is simple, they are cheaper because they offer inconsistent performance - pretty basic concept. Personally, I would not spend $800+ on ATI when I don't get the performance return out of it. What's the point? I might as well spend the extra $300 since I like Folding@Home, and I play a lot of games that scale well on Nvidia cards, like Dead Space for example. So why would I buy ATI? The answer is I wouldn't. I want a GPU that I can do more with than just gaming, and Nvidia has proven to have a ton more uses for their GPUs than ATI, as Frank said above. Sure ATI can fold, but nowhere near the level that Nvidia can, so it's a pretty cut and dried choice for me. I want the company that busts their asses to give you maximum performance without having to wait a month or two at a time for a driver update, if not longer, to add support for the game that you want to play now. To me, that is completely unacceptable, and this has been a complaint for a very, very long time from communities all over the Internet when it comes to CrossFire scaling. Who knows the reason for this, but they need to start working on their drivers to be more efficient, because in some games it's just embarrassing, and it has to be frustrating to owners.

If you are running a CrossFire motherboard, what do you do? Simple - either buy a high-end Nvidia card such as the GTX 295, or run CrossFire. You may not get the same performance that you would with SLI, but you will still get an upgrade nonetheless. ATI cards are decent in certain games, no doubt about that, they are just not consistent across the board like I said earlier. My suggestion is, if you plan on doing a multi-GPU setup, do your research. Look at the games you want to play, and see how they scale. See where Nvidia and ATI are performance-wise, and make your purchase based on what you learn from reading and asking questions.

The idea of this article was to give people an idea as to where everything stands with different GPU setups. At the end of the day, you the user will have to decide exactly where your hard earned money is going to go. Our goal is to try and keep these types of articles coming for you as newer cards come out, so you can make good choices - because you can bet there are going to be more games like Crysis coming.

I know people will call me a fanboy, but it's my money. My Nvidia rigs are way more consistent than people's CrossFire rigs, and I get a lot more Folding@Home work units completed than the majority of people - and that's what matters to me.

As for me, my choice is a given - I have Green in my blood!