review a brief look at three way sli

A brief look at three-way SLI

A funny thing happened after our Radeon HD 3870 X2 review. I almost immediately transitioned over to working on our review of Skulltrail, Intel’s eight-core monstrosity that we overclocked to 4GHz. Still woozy from Skulltrail’s exhaust fumes, I then decided to turn my attention to yet another ridiculously expensive and powerful gizmo: three-way SLI.

At this point, my editor, if I had one, would probably be clobbering me. No doubt I could more profitably be spending my time reviewing hardware that most folks might actually, you know, want to purchase. And heck, I’m getting really close on that review of the 45nm Core 2 Duos—honest. But I couldn’t resist a brief detour involving a howling phalanx of GeForce 8800 Ultras, mostly because I wanted to see whether we’d finally found the hardware equal to the task of making Crysis run smoothly at high resolutions and quality levels.

So what we have here is a quick foray into the warm and sordid world of three-way SLI. We’ll take a look at how a trio of Ultras performs in a number of games, with a little extra emphasis on the system killer du jour.

Our three-way SLI test rig on the bench

What it takes to make a trio sing
Building a dual-GPU rig these days isn’t exactly rocket science, let alone shotgun proteomics. You just have to select the right graphics cards, a compatible dual-slot motherboard, and a power supply that’s up to the task. Generally, assembling the setup beyond that isn’t any more difficult than building any other PC.

If you wish to add a third graphics card, though, matters become much more complicated in a hurry. Turns out that going up a weight class is no trivial—or inexpensive—matter.

First of all, Nvidia only makes two graphics cards capable of running in a three-way configuration, the GeForce 8800 GTX and its up-clocked brother, the 8800 Ultra. Both of these cards have dual SLI connectors onboard, and no other GeForce 8-series graphics card does. Since an 8800 Ultra can cost somewhere north of 650 bucks, investing in three is a substantial commitment.

Next, you’re going to need a motherboard with three PCIe x16 slots and Nvidia’s blessing. At present, that approval is limited to mobos based on the nForce 680i SLI and 780i SLI chipsets. You’ll also need a fast Core 2 processor to populate that board’s CPU socket. Nvidia recommends something in the 3GHz range with at least two cores, like the Core 2 Extreme X6800. Four cores would, of course, be even better.

Once you have the necessary cards in the necessary slots, you’ll need to connect them all together, and for that, you’ll need a custom SLI connector doober that looks like this:

Regular two-way SLI connectors ship with most SLI-ready motherboards, but this puppy is harder to come by. Nvidia has links to some places where you can order them online on its website. Prices appear to range between 10 and 20 bucks. That seems like an awfully lot for a connector doohickey, but hey, suck it up. If you let that extra expense put you off, you were never really a good candidate for three-way SLI anyhow.

Besides, there are much larger matters at hand, like the power supply question. When I was assembling our test system, I figured any one of the several thousand-watt beasts we have kicking around in Damage Labs would surely suffice. Turns out I was wrong. Nvidia recommends at least a 1100W PSU, and it will need to have a total of six 6-pin PCIe auxiliary power connectors onboard—two each for the three graphics cards.

This oughta do

In order to get the right connector count, we turned to this PC Power & Cooling Turbo-Cool 1200, the big brother of the model that delivered beautifully clean and efficient power in our recent PSU roundup. The Turbo-Cool 1200 handled our three-way SLI rig, no problem. Like its smaller counterpart, though, this PSU certainly makes its presence known with quite a bit of fan noise, even at idle. Such is the price of prodigious power.

Oh, yes. You’re going to need a big, roomy enclosure to house all of this stuff and keep it cool. Nvidia recommends a Cooler Master Cosmos or a Silverstone TJ10. You may want to cut an extra hole or two in the side of the case and mount fans in them for additional cooling, as well.

Finally, you’ll need to install Windows Vista on that new system in order to get three-way SLI working. Windows XP isn’t supported. Nvidia also strongly recommends installing this Vista hotfix in order to ensure SLI performance scaling. That hotfix is one of five Vista updates they currently recommend.

How to harness a howling phalanx
What you get for your trouble, of course, is one of the most astoundingly powerful GPU configurations anywhere, with a total of 2.25GB of GDDR3 memory dedicated to video RAM and over 1.7 teraflops of raw shader power. In fact, don’t even bother going for a three-way system unless you’re going to hook it up to at least a 30″ display with something like 2560×1600 resolution. Anything less would be a waste in most of today’s games. You’re going to want to push display resolutions and quality levels to the max in order to make the most of three-way SLI.

Those of you who remember Quad SLI will know why I say that. Quad SLI had difficulty delivering consistent performance gains over two-way SLI, especially in DirectX 9, where a three-buffer limitation hampered load-balancing efforts. Nvidia argues that three-way SLI avoids this limitation and is more broadly compatible, and they’re probably right. Still, three-way SLI will almost surely need to be pushed in order to show its true potential.

For this quick look at three-way SLI, we’ve tested at a range of resolutions up to 2560×1600, starting with a focus on 16X anisotropic filtering and 4X antialiasing where possible. Rather than push on to extensive testing at higher AA levels, we then largely concentrated our efforts on Crysis, the one game that really seems to need the sort of graphical oomph a three-way SLI rig can provide.

Our testing methods
As ever, we did our best to deliver clean benchmark numbers. Tests were run at least three times, and the results were averaged.

Our test systems were configured like so:

Processor Core
2 Extreme X6800
2 Extreme X6800
(266MHz quad-pumped)
(266MHz quad-pumped)
Motherboard Gigabyte
nForce 680i SLI
F7 P31
680i SLI SPP
ICH9R nForce
680i SLI MCP

Matrix Storage Manager 7.8

(4 DIMMs)
(4 DIMMs)
x Corsair
at 800MHz
x Corsair
at 800MHz
latency (CL)
4 4
to CAS delay (tRCD)
4 4
precharge (tRP)
4 4
time (tRAS)
18 18
2T 2T
Audio Integrated
with RealTek drivers
nForce 680i SLI/ALC850
with RealTek drivers
Graphics Dual

Radeon HD 3870 512MB PCIe

with 8.451.2-080116a-057935E drivers
8800 GT 512MB PCIe

with ForceWare 169.28 drivers

Radeon HD 3870 512MB PCIe

with 8.451.2-080116a-057935E drivers
8800 Ultra 768MB PCIe

with ForceWare 169.28 drivers

Radeon HD 3870 X2 1GB PCIe

with 8.451.2-080116a-057935E drivers
8800 Ultra 768MB PCIe

with ForceWare 169.28 drivers
8800 GT 512MB PCIe

with ForceWare 169.28 drivers
GeForce 8800 GTS 512MB PCIe

with ForceWare 169.28 drivers
8800 Ultra 768MB PCIe

with ForceWare 169.28 drivers
Caviar SE16 320GB SATA
OS Windows
Vista Ultimate
x86 Edition
KB936710, KB938194, KB938979,
KB940105, KB945149,
DirectX November 2007 Update

Thanks to Corsair for providing us with memory for our testing. Their quality, service, and support are easily superior to no-name DIMMs.

Except for the three-way SLI rig, our test systems were powered by PC Power & Cooling Silencer 750W power supply units. The Silencer 750W was a runaway Editor’s Choice winner in our epic 11-way power supply roundup, so it seemed like a fitting choice for our test rigs. Thanks to OCZ for providing these units for our use in testing.

Unless otherwise specified, image quality settings for the graphics cards were left at the control panel defaults. Vertical refresh sync (vsync) was disabled for all tests.

We used the following versions of our test applications:

The tests and methods we employ are generally publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.

The theory of three
We’ve chosen to build our three-way SLI rig with GeForce 8800 Ultras because, hey, if you’re gonna go all out, you might as well go all out. The Ultra is a very powerful graphics card in itself, but here’s what happens, in theory, if you put three of them together.

fill rate

Peak bilinear

Peak bilinear
FP16 texel


GeForce 8800 GT 9.6 33.6 16.8 57.6 504
GeForce 8800 GTS 10.0 12.0 12.0 64.0 346
GeForce 8800 GTS 512 10.4 41.6 20.8 62.1 624

GeForce 8800 GTX

13.8 18.4 18.4 86.4 518
GeForce 8800 Ultra 14.7 19.6 19.6 103.7 576
GeForce 8800 Ultra 2-way SLI 29.4 39.2 39.2 207.4 1152
GeForce 8800 Ultra 3-way SLI 44.1 58.8 58.8 311.0 1728
Radeon HD 2900 XT 11.9 11.9 11.9 105.6 475
Radeon HD 3850 10.7 10.7 10.7 53.1 429
Radeon HD 3870 12.4 12.4 12.4 72.0 496
Radeon HD 3870 X2 26.4 26.4 26.4 115.2 1056

The numbers are staggering. 311 GB/s of memory bandwidth, more than 1.7 teraflops of shader arithmetic, and nearly 60 Gtexels per second of FP16 texture filtering capacity—almost five and a half times that of a GeForce 8800 GT. Of course, extracting the full potential from three graphics cards is much more difficult than it is with one graphics card, but still… wow.

Our three-way rig leads all contenders in 3DMark’s synthetic tests of GPU power, with the lone exception of the simple vertex shader test, where it seems to run into a scaling issue. In terms of delivered multitextured fill rate, the three-way SLI system roughly doubles the Radeon HD 3870 X2, AMD’s current fastest graphics solution.

Call of Duty 4: Modern Warfare
We tested Call of Duty 4 by recording a custom demo of a multiplayer gaming session and playing it back using the game’s timedemo capability. Since this is a high-end graphics card we’re testing, we enabled 4X antialiasing and 16X anisotropic filtering and turned up the game’s texture and image quality settings to their limits.

Thanks to a recent change to Nvidia’s drivers, we were (finally!) able to test at 1680×1050, even though that’s not our display’s native resolution. Consequently, we’ve chosen to test at 1680×1050, 1920×1200, and 2560×1600—resolutions of roughly two, three, and four megapixels—to see how performance scales.

Now you can see why three-way SLI is billed as a solution only for those with large displays. At lower resolutions, three-way is slower than a dual-GPU config. As the resolution rises, though, its performance remains steady, until it takes the lead at 2560×1600. Notice, however, that two 8800 Ultras are still pumping out over 70 frames per second at the top resolution.

Half-Life 2: Episode Two
We used a custom-recorded timedemo for this game, as well. We tested Episode Two with the in-game image quality options cranked, with 4X AA and 16 anisotropic filtering. HDR lighting and motion blur were both enabled.

This game doesn’t work out so well for the three-way SLI setup. Even at four megapixels, the dual-GPU system proves just as fast, and at 1920×1200, two 8800 GTs are nearly as quick. Those two 8800 GTs also deliver a very playable 60 frames per second at 2560×1600.

Since several GPU configs can run Episode Two pretty well at this resolution, and since we hadn’t had much success with three-way SLI scaling at 4X AA, I thought this might be a good opportunity to test at higher AA levels. I ran a quick test at 2560×1600 with two-way and three-way SLI using Nvidia’s 8X and 16X coverage sampled AA modes.

Well, it was worth shot. If you can’t quite tell, those two lines are overlapping. Turns out you can run either setup at 8X or 16X CSAA without any performance penalty.

Enemy Territory: Quake Wars
We tested this game with 4X antialiasing and 16X anisotropic filtering enabled, along with “high” settings for all of the game’s quality options except “Shader level” which was set to “Ultra.” We left the diffuse, bump, and specular texture quality settings at their default levels, though. Shadows, soft particles, and smooth foliage were enabled. Again, we used a custom timedemo recorded for use in this review.

The results here look very similar to what we saw in Call of Duty 4. Three-way SLI has some overhead that makes it a little slower at lower resolutions, but it just doesn’t slow down as the number of pixels onscreen rises.

Unreal Tournament 3
We tested UT3 by playing a deathmatch against some bots and recording frame rates during 60-second gameplay sessions using FRAPS. This method has the advantage of duplicating real gameplay, but it comes at the expense of precise repeatability. We believe five sample sessions are sufficient to get reasonably consistent and trustworthy results. In addition to average frame rates, we’ve included the low frames rates, because those tend to reflect the user experience in performance-critical situations. In order to diminish the effect of outliers, we’ve reported the median of the five low frame rates we encountered.

Because UT3 doesn’t support multisampled antialiasing, we tested without AA. Instead, we just cranked up the resolution to 2560×1600 and turned up the game’s quality sliders to the max. I also disabled the game’s frame rate cap before testing.

This is a case where three-way’s overhead just isn’t worth it, mainly because UT3 doesn’t support antialiasing. Nvidia does have an option in its control panel to force on antialiasing. When I tried it, though, it exacted a big performance hit with 4X AA. Playability was borderline even with three-way SLI. However, I didn’t spend too much time on it, because I was concentrating on….

We tested Crysis in several different ways, starting with the game’s “high” quality options and the GPU benchmark script Crytek supplies with the game.

Please note that all of the results you see below for the Radeons come from a newer graphics driver, version 8.451-2-080123a, than the ones we used for the rest of our tests. This newer driver improved Crysis performance noticeably over the older one, both in benchmarks and when playing the game.

Testing this way, we see only a minimal difference between three-way SLI and two-way. Three-way does exhibit its usual pattern of not slowing down as the screen resolution increases, amusingly showing a slight upward bend in the scaling results when we go to 1920×1200, as if it were begging for more.

I was frustrated, however, about the fact that we couldn’t seem to reach above about 40 FPS, almost no matter what. Thinking perhaps Crysis was CPU-bound, I decided to swap out our Core 2 Extreme X6800 processor for a Core 2 Extreme QX6850. With another 66MHz of clock speed, two more cores, and a faster 1333MHz bus, I figured the QX6850 might help improve performance.

No such luck. Frame rates remained at about 40 FPS, even with the QX6850. Despite the hype, Crysis doesn’t appear to benefit from more than two cores. I would have gone for an even faster CPU, like a Core 2 Extreme QX9650, but our nForce 680i SLI motherboard doesn’t support 45nm processors. Heck, I would have tested with our Skulltrail rig overclocked to 4GHz, but Nvidia put the kibosh on that one.

I was a little dubious about Crytek’s built-in GPU benchmark at this point—it’s a flyover that covers a lot of ground quickly and appears to stream in lots of data in a short period, possibly making it I/O bound—so I decided to see what I could learn by testing with FRAPS instead. I chose to test in the “Recovery” level, early in the game, using our standard FRAPS testing procedure (five sessions of 60 seconds each). The area where I tested included some forest, a village, a roadside, and some water—a good mix of the game’s usual environments.

And we’re still not seeing much in the way of three-way SLI performance scaling or frame rates over 40 FPS. In the interests of teasing out some SLI scaling differences—and of pure, sweet science—I decided to try testing on Crysis‘ “very high” settings with 4X antialiasing, as well.

Here, three-way SLI is clearly the least unplayable of all four unplayable configurations tested. Victory!

Power consumption
We measured total system power consumption at the wall socket using an Extech power analyzer model 380803. The monitor was plugged into a separate outlet, so its power draw was not part of our measurement. The cards were plugged into a motherboard on an open test bench.

The idle measurements were taken at the Windows Vista desktop with the Aero theme enabled. The cards were tested under load running UT3 at 2560×1600 resolution, using the same settings we did for performance testing.

Note that the SLI configs were, by necessity, tested on a different motherboard than the single cards, as noted in our testing methods section. Also, because it required extra help, the three-way SLI system was tested with the PC Power & Cooling Turbo-Cool 1200 power supply. The other systems were tested with a Silencer 750W PSU.

We have breached the 600W mark! Whoa. That’s like 26 swirly bulbs of “60W equivalent” output. Obviously, a three-way SLI rig will spin the meter. Even at idle, it draws over 400W.

Noise levels
We measured noise levels on our test systems, sitting on an open test bench, using an Extech model 407727 digital sound level meter. The meter was mounted on a tripod approximately 12″ from the test system at a height even with the top of the video card. We used the OSHA-standard weighting and speed for these measurements.

You can think of these noise level measurements much like our system power consumption tests, because the entire systems’ noise levels were measured, including the stock Intel cooler we used to cool the CPU. Of course, noise levels will vary greatly in the real world along with the acoustic properties of the PC enclosure used, whether the enclosure provides adequate cooling to avoid a card’s highest fan speeds, placement of the enclosure in the room, and a whole range of other variables. These results should give a reasonably good picture of comparative fan noise, though.

Unfortunately—or, rather, quite fortunately—I wasn’t able to reliably measure noise levels for most of these systems at idle. Our test systems keep getting quieter with the addition of new power supply units and new motherboards with passive cooling and the like, as do the video cards themselves. I decided this time around that our test rigs at idle are too close to the sensitivity floor for our sound level meter, so I only measured noise levels under load.

I did, however, make an exception for the three-way SLI system with its 1200W PSU, which creates enough noise to register 52.6 dB on the meter at idle, even though the rest of the systems were at or under 40 dB.

The three-way system was also the noisiest of the bunch when running a game. Some of that noise comes from its relatively loud PSU, but subjectively, the coolers on the three 8800 Ultras sandwiched together certainly seemed to be contributing to the cacophany, as well.

GPU temperatures
Per your requests, I’ve added GPU temperature readings to our results. I captured these using AMD’s Catalyst Control Center and Nvidia’s nTune Monitor, so we’re basically relying on the cards to report their temperatures properly. In the case of multi-GPU configs, well, I only got one number out of CCC. I used the highest of the numbers from the Nvidia monitoring app. These temperatures were recorded while running UT3 in a window.

Cramming three of these cards together raises the temperature quite a bit, which helps explain why the coolers generate more noise.

What can I say? The picture isn’t terribly rosy. In terms of performance, what we saw out of three-way SLI was more or less what we expected: that you’ll need a 2560×1600 display in order to really realize the benefits of three GPUs. Even then, in none of the games we tested did we see a case where going from two cards to three made a real difference in terms of playability. (Going from 80 FPS to 95 FPS doesn’t count.) The closest we came to that may have been Crysis, but that game seems to have its own performance scaling problems—at least at the “high” and “very high” settings—that involve factors other than the graphics subsystem alone. Perhaps, in some games, turning up antialiasing levels to 8X or 16X will help tease out some of the benefits of having a third GPU. That wasn’t the case when we tried it in Half-Life 2: Episode Two, however.

Subjectively, playing games on a three-way rig wasn’t especially exciting, either. The main difference I noticed was the noise emanating from the system, a reality we quantified with our power, noise, and temperature measurements. Three-way SLI truly is an extreme solution, and it comes with all of the drawbacks that such things tend to have. I’m sure it’s possible with careful component selection, artful thermal design, and probably some form of water cooling to make a three-way SLI system that isn’t especially loud. But it will draw lots of power and expend it into the room as heat, almost inescapably.

Personally, I don’t think it’s worth bothering, and I haven’t even mentioned the expense involved yet. Two GeForce 8800 GTs should be up to the task of driving a 30″ wide-aspect display reasonably well in most games, and a pair of those costs less than a single 8800 Ultra. If that’s not enough for you, in my view, the very best config Nvidia has to offer right now is a pair of GeForce 8800 GTS 512s. This card is, like the 8800 GT, based on the new G92 GPU. As a result, the GTS 512 has H.264 decode acceleration for HD video playback, and it has more texture filtering and shader arithmetic power than the 8800 Ultra.

Having said all of that, I’m still glad that three-way SLI exists, simply because it expands the horizons a little bit. In fact, I’d like to see more-than-two-way setups with G92 GPUs become available soon, since they’d be more attractive overall. So long as a system like this is possible, somebody out there is going to be looking for ways to take full advantage of it. Crytek seems to be close. If and when they get there, it should be a glorious sight to behold.

Scott Wasson

Latest News

pancakeswap (CAKE)
Blog, Cryptocurrency, Investments, News, Price Prediction

PancakeSwap Price Is Up 33% in 30 Days as February Begins – Here’s Why You Need to Buy Now!

eco crypto
Blog, Cryptocurrency, Investments

The 9 Best New Cryptocurrencies YOU NEED TO BUY in 2023

With hundreds of cryptocurrencies created daily, it can be tough to differentiate the projects that won’t last longer than a few months from those  that will explode in value. Of...

3.1 m memag
blockchain, Blog, Cryptocurrency, Gaming, Innovation, News

Meta Masters Guild (MEMAG) Hits 3.4 Million – Buy Now as Phase 6 Begins!

News about the first mobile-based Web3 gaming guild has spread quickly among the crypto community and the latest numbers back up the hype. Even though the presale started just a...

crypto reddit
Blog, Cryptocurrency, Investments, News, Price Prediction

7 Best Cryptocurrencies On Reddit To Invest In 2023

Blog, Cryptocurrency, Investments, Sustainability

The Clock is Ticking to Buy this Eco-Crypto – 7 Reasons C+Charge Will Explode Post-Presale!

savannah marshall fght
Blog, Cryptocurrency, News

Boxing World Champ, Savannah Marshall Partners with FGHT as presale approaches to $4 million!

Twitter Files

Elon Musk’s Twitter Pledges Free Write-Only API Access To Bots Creating “Good” Content