Home Archive Search Forum Reviews IRC Chat Shop


Search Site
NVIDIA Stuff
Executive Profiles
NVIDIA Drivers
Laptop Drivers
Beta Drivers
Archived Drivers
Driver Feedback
GPU Computing
OpenCL Computing
Direct Compute
Desktop Products
Workstation GPUs
Desktop GPUs
Laptop GPUs
Netbook GPUs
Handheld Devices
Portable Media
Automotive Devices
Server Solutions
Application Engines
Apple Products
Game Consoles
System Tools
Power Packs
Get A Balanced PC
Pure Video SD
Pure Video HD
Extreme HD
GeForce PC Kit
NVIDIA 3D Vision
NVIDIA Cool Stuff
NVIDIA Software
NVIDIA PhysX
NVIDIA CUDA Zone
GPU Venture Zone
NVIDIA nZone
NVIDIA SLI Zone
SLI App Request
SLI Profile Patches
Developer Zone
NVIDIA Support
FreeBSD Support
Linux Support
Solaris Support
NVISION '08
GPU Conference '09
NVIDIA at CES '10
PAX East 2010 1
NVIDIA nTersect
NVIDIA Newsroom
NVIDIA at Facebook
NVIDIA at Flickr
NVIDIA at Twitter
NVIDIA at YouTube
NVCUDA at YouTube
NVIDIA Online Store
1 March 26, 2010
EVGA Stuff
EVGA E-LEET
EVGA Precision
GPU Voltage Tuner
SLI Enhancement
EVGA Gear
Reviews and Awards
Articles
GeForce GTX 295
GeForce GTX 280
GeForce GTX 260
GeForce GT 240
GeForce 9800 GTX
GeForce 9800 GX2
GeForce 9600 GT
GeForce 8800 Ultra
GeForce 8800 GTX
GeForce 8800 GTS
GeForce 8800 GT
GeForce 8600 GTS
GeForce 8500 GT
GeForce 7950 GX2
GeForce 7950 GT
GeForce 7900 GTX
GeForce 7900 GS
GeForce 7800 GTX
Watercooling Project
My Book 500GB
Raptor Hard Drive
Guide To Doom 3
Other Stuff
Game Releases
  By Date
  Alphabetical
Litigation
  FTC vs. Intel Corp.
Steam
  Hardware Survey
CES 2010
  Press Conference
GF100 White Papers
  GPU Architecture
  GF100 Compute

MSI GeForce 6600 GT SLI Review - Page 3 of 5

GAME BENCHMARKS

UNREAL TOURNAMENT 2004

May as well get the bad news out of the way first. Try as I might, I simply could not get SLI to work properly in any mode with this game. Combine comparatively immature drivers for the technology with this game's major CPU limitations and what you see below is what you get. For this test I used the Primeval timedemo, which demos an intense Onslaught match with several players, vehicles and explosions. I ran the benchmark three times at each setting and took the average frame rate. All in-game settings were put to their maximum, except physics detail, which was set to low to try to lower CPU overhead. Sound also was disabled.

Unreal Tournament 2004 - Primeval Demo

You will notice how I used the split-frame rendering mode to obtain my results. The default profile in the 71.81 drivers is for alternate-frame rendering. I was reading Chris Ray's SLI investigation in the forums. He realized a performance increase in the most GPU-intensive situations in UT 2004 by switching the profile to SFR. Obviously this didn't work for me and in fact had quite the opposite reaction.

Using the default AFR profile made no difference in terms of benchmark scores. The differences were often less than 1 frame per second, even at the highest resolution and image quality settings. I wouldn't say the picture is totally bleak, however. In addition to other game tests showing improvements with SLI, you can use the NvSLI Application Optimizer to force UT 2004 to use single GPU rendering even if you have the SLI checkbox enabled in the driver control panel. Set it, reboot and forget it. Scores jump back up to their normal single GPU levels.

HALF-LIFE 2

I will spare you the particulars of everything this game has going for it. In spite of a rocky launch, Valve Software has been, in my opinion, pretty responsive to fixes, bugs, and performance issues relating to the single-player aspect of this game. I personally don't play any of the multiplayer aspects, but I continue to enjoy the solo campaign.

In this game, I used the popular benchmarking program Fraps to record game performance in a self-created demo. I chose a nice segment of the "Highway 17" level that features a variety of indoor and outdoor scenery. I spent time calculating the best ways to take my run, and then ran through my own demo three times using Fraps' built-in benchmarking capability. All in-game settings were set to the game defaults, which was high except for the reflection setting, which defaulted to "Reflect world."

Half-Life 2 - Highway 17

Some nice improvements here when you up the resolution from 1024 to 1280 and from 1280 to 1600. I highlighted the 1600x1200 results to illustrate this point. While 4xAA/8xAF remains unplayable, that's more a limitation of the 128 MB of video memory and 128-bit memory interface. Imagine the results were I running a pair of PCI-Express 6800 GT or Ultra cards.

Out of sheer curiosity, I decided to run a quick test on this map at 1280 resolution but dropping the AA and AF to 2x and 4x, respectively, with SLI enabled. Scores from the averaged out runs: 35 min and 82 avg. The 35 happened only for a spilt second and was not to be seen in any other parts of the test. I personally find this setting ideal for my tastes - solid IQ and performance all rolled into one, no doubt thanks to the second video card.

Far Cry

Here's the first game that was engineered to take advantage of the GeForce 6-series support for Shader Model 3.0. While it was through a patch, it was nice to see developer hop on-board with this added feature. Tests of this game used the split-frame rendering mode. As with Half-Life 2, I used Fraps to record a demo. I used a portion of the Training level, again featuring indoor and outdoor environments. All in-game options were set to their highest levels, except for shadows, which were set on medium. Hardware audio was disabled as well as it would cause the game to stutter and slow to a crawl.

Far Cry - Training

SLI also is doing its job in this title, albeit somewhat less effectively than in Half-Life 2. This likely is due, however, to some limitations within the game engine. Not only in this level but also other spots in the game give problems for video cards. My former GeForce 6800 GT would occasionally struggle in some areas, that's why you're seeing a bit more consistency in the low frame rate measures at 1024 and 1280.

Using SLI at 1280x1024 and dropping the anti-aliasing to 2x (low) gave scores of 27 min and 45 avg.

Next Page: Doom 3, Halo, and Overclocking Performance

Last Updated on March 16, 2005


Table of Contents

Shop Online at PriceGrabber

nV News - Copyright © 1998-2011. All rights reserved.
Reproduction in any form or medium without written permission of the site's owner is prohibited.
Search Products
Search
for


Graphics Utilities
AMD GPU Clock Tool
ATITool
aTuner
EVGA Precision
EVGA Voltage Tuner
Gainward ExperTool
GPU-Shark
GPU Voltage Tuner
Fraps
FurMark
GLview
GPU Caps Viewer
GPU Shark
GPU-Z
MSI Afterburner
nHancer
NiBiTor
NVClock (Linux)
NvTempLogger
NVTray
PowerStrip
RivaTuner
SLI Profile Tool
3DCenter Filter Test
3DMark Vantage
Add-In Partners
Albatron
ASUS
AXLE
BFG Technologies
BIOSTAR
Chaintech
Colorful
ELSA
EVGA
GAINWARD
GALAXY
GIGABYTE
FORSA
FOXCONN
Inno3D
Jaton
Leadtek
MSI
Palit
PNY
Point of View
Prolink
SPARKLE
XFX
ZOGIS
ZOTAC
For Developers
ACM SIGGRAPH
AMD
DevMaster.net
flipCode
Gamasutra
GameDev.net
GPGPU
Intel
Microsoft
CiteSeer
NeHe Productions
NVIDIA
OpenGL.org
Programmers Heaven
Real-Time Rendering
Stanford Graphics
3dRender.com
Associates
Benchmark Reviews
Fraps
GeForce Italia
GPU Review
Hardware Pacers
LaptopVideo2Go
MVKTECH
News3D (NVITALIA)
OutoftheBoxMods
OSNN.net
Overclocker Cafe
PC Extreme
PC Gaming Standards
PhysX Links & Info
TestSeek
3DChip (German)
8Dimensional