MaximumPC 2006 03

A

long with all the emails I get promising richer, thicker hair or

entreating me to help the crown prince of Nigeria, I regularly

receive messages making odd tech claims. One that recent-

ly perked my interest stated that removing the bridge between two

SLI GeForce cards actually increases performance.

How can this be? It wasn’t until recently that new drivers even

allowed us to run SLI without a bridge connector, so how was a

performance improvement possible? Could the conspiracy theo-

ries be correct? Is nVidia forcing people to buy bridge connectors

because the company also secretly owns the small firm that makes

those parts too?

We decided to get to the bottom of the mystery. As you know,

SLI works by connecting two graphics cards together to share the

workload of image rendering. The cards can communicate through

the PCI Express bus, but they also communicate via a special

bridge connector. Or so we’re told by nVidia.

After configuring an Asus A8N-SLI Deluxe motherboard with an

Athlon 64 FX-55, 2GB of DDR400, and two eVGA GeForce 7800

GTX 512MB cards, we ran the rig through a series of benchmarks,

both with and without the bridge in place.

The verdict? Don’t run your rig without the bridge card. We

found that pulling the little connector seriously impedes perfor-

mance in every benchmark we ran. Sure, the videocards still work,

but not at their maximum speed.

The nVidia eggheads weren’t surprised. Without the bridge con-

nector, the cards exclusively rely on the PCI Express bus to com-

municate. And while the low-end cards—such as the GeForce 6600

and 6600 LE—only use the PCI-E bus to communicate, the bus

doesn’t provide enough bandwidth for the high-end cards. As you

do more 3D work, you generate more traffic, which can swamp the

bus. When you crank up a pair of 512MB 7800 GTX cards to high

resolutions and maximum AA settings, you need the dedicated link

to keep the cards from overloading the PCI-E bus.

A

lice Nelson and Florence Johnston would agree: Cleaning the

remnants of used thermal pads and thermal grease from a CPU

cooler is a bitch. And I concur: As someone who has done his fair

share of CPU showdowns, I can tell you that it’s no fun to try to

clean that funky stuff up.

We normally keep a large bottle of 99-percent isopropyl alcohol

handy for the job, but sometimes that isn’t even enough. When I saw

Arctic Silver’s ArctiClean ($7, http://www.arcticsilver.com ), I decided to give it

a whirl. ArctiClean uses a two-step process: The citrus-and-soy-based

“thermal material remover” wipes out most of the gunk (including wax

thermal pads) and the follow-up “thermal surface purifier” removes the

cleaning residue from the

heatsink or CPU.

To see how well it worked,

I found an old heatsink in the

Lab that had silver-based

thermal compound dried and

caked on it. On one half I used

the ArctiClean combo, and

on the other side I used stan-

dard alcohol. The results: pretty impressive. While I can’t say that the

ArctiClean actually cleans the surface better than alcohol (to the naked

eye anyway), I can say it takes a lot less elbow grease. Only with a lot

more scrubbing and buffing, and several extra passes was I able to get

the alcohol side to look as shiny as the ArctiClean side. Arctic Silver

claims that ArctiClean contains flash corrosion inhibitors to reduce the

corrosion layer on copper or aluminum, but I couldn’t test that.

What I can tell you is that ArctiClean makes cleaning easier and

faster, and gets you a spic-and-span heatsink surface in less time than

standard alcohol.

Gordon Mah Ung

Kisses His Iso propyl

Goodbye

Think you can improve performance by removing

your SLI connector? Think again

New ArctiClean cleans better than alcohol

58 MA XIMUMPC MARCH 2006

in the lab REAL-WORLD TESTING: RESULTS. ANALYSIS. RECOMMENDATIONS

The left side was cleaned with

ArctiClean; the right with alchohol.

We recommend you run your two videocards with the SLI connector.

MICHAEL BROWN

Bats down

the SLI Bridge

Rumor

BENCHMARKS

Best scores are bolded. All tests were run at 1600x1200 on our standard videocard test bed, with

two 512MB eVGA eGeForce 7800 GTXs.

WITHOUT BRIDGE 25,271 12,140 75.3 117.8

WITH BRIDGE 32,094 12,526 99.2fps 153.3fps

3DMARK03 3DMARK05 DOOM 3 FAR CRY