***The Ultimate SLI Test***

***The Ultimate SLI Test***

http://www.slizone.com/docs/IO/27574...cott_Sarbe.jpg

Introduction

Many people have been wondering about a few questions regarding SLI and required cabling. It has pretty well known that people use an SLI cable in order to run SLI setups but few have ever wondered about whether SLI would even work without a cable. Apparently there have been some sources that claim this is possible and others that deny it. In an effort to provide everyone with factual up-to-date information on it's possibilities. This is exactly the test I will be conducting for everyone.

I have also been noticing a few people running SLI using a Tri-SLI cable (dual connectors) and I was interested to see if there is any chance that running such a setup will increase the performance any. Therefore I will also be testing on whether it makes any difference to use the Tri-SLI cable on a regular SLI setup as theoretically it should increase the performance due to both connectors being in use simultaneously, but it is of course possible that NVIDIA never implemented that ability, therefore I am here to answer all these questions and hopefully anymore that may arise will also be answered.

Testing will include a 3DMark Vantage Professional Edition benchmark run with a Single, SLI (with no cable), SLI (with SLI cable), and SLI (with Tri-SLI cable) test. All test will be conducted using the following system.

Test Setup

Motherboard: EVGA nForce 780i
BIOS Version: P09 (Latest)
CPU: Intel Core 2 Duo E8500 (4.27 GHz)
RAM: OCZ Vista Upgrade 4GB 800 MHz DDR2 (5-6-6-18-2T Timings)
GPU: XFX GeForce 9800GTX (Single + SLI)
HD: Samsung 80GB SATA
PSU: Delta Electronics 800 Watt
CPU Heatsink: Xigmatek Dark Knight HDT-1283V
OS: Microsoft Windows 7 Ultimate 64-Bit
GPU Drivers: NVIDIA ForceWare 186.18 64-Bit

Benchmark Settings

3DMark Vantage Professional Edition: Default Performance Settings

Testing and Results

Single

http://img42.imageshack.us/img42/7494/dscn2093d.jpg

http://img705.imageshack.us/img705/7728/singleq.jpg

SLI (with no cable)

http://img705.imageshack.us/img705/7671/dscn2094.jpg

http://img42.imageshack.us/img42/1702/slinocable.jpg

SLI (with SLI cable)

http://img709.imageshack.us/img709/8167/dscn2096.jpg

http://img705.imageshack.us/img705/369/slislicable.jpg

SLI (with Tri-SLI cable)

http://img709.imageshack.us/img709/6295/dscn2097e.jpg

http://img705.imageshack.us/img705/5...rislicable.jpg

Conclusion

After conducting these tests I was able to deduce that apparently using the cable does in fact make a difference but a very minor one. Obviously compared to a single card there is a huge difference but that's expected. Adding the single cable on gave a minor boost in performance and changing the Tri-SLI one gave no performance increase at all. This means that the Tri-SLI is absolutely useless in an SLI setup. As a summary, I was quite surprised that the cables made such a minor difference but apparently they aren't as important as one would think.

Thanks for the science! Funny how you only got 200 points with the cable connected. I didn't think the tri cable would make any difference, besides maybe adding some more structural integrity to the setup. I wonder if these findings translate to the 200 series.

Very good testing indeed! The SLI bridge is clearly just to split the bandwidth communications between the GPU's. Or moreso a more efficient communication path. Looks like your CPU score is down a bit on the second test, which would bump up the test with SLI bridge a bit, but overall SLI bridge only improved ~1% in average framerate.

What would be interesting is to test both cards on a PCI-E 1.x vs PCI-E 2.x as the bandwidth of course is basically double on PCI-E 2.x.

Thanks for posting!