NVIDIA nForce 4 SLI: Intel 3.73GHz Extreme Edition, 2 x 512MB Corsair TWIN2X PC5400, 2 x NVIDIA 6800GT, 73GB WD Raptor, Windows XP SP1.

VIA PT894: Intel 3.73GHz Extreme Edition, 2 x 512MB Corsair TWINX PC4400, MSI NX6800GT, 73GB WD Raptor, Windows XP SP1.

EPoX EP-5LWA+ 925XE: Intel 3.73GHz Extreme Edition, 2 x 512MB Corsair TWIN2X PC5400, MSI NX6800GT, 73GB WD Raptor, Windows XP SP1.

Going up against the NVIDIA nForce 4 SLI will be the EPoX EP-5LWA+ (a 925XE based board), and the reference VIA PT894 PT Series. The setups all share similar peripheral components, except the PT Series will be running Corsair DDR ram rather than DDR2 since our board did not support it.

Onboard audio was enabled in the BIOS for all the boards, but not used during game testing. All tests on the nForce 4 board was done with SLI enabled. Any system tweaks and ram timings were configured to the best possible for each platform. All benchmarks will be run a total of three times with the average scores being displayed.

Test Software is as follows:

SiSoft Sandra 2005 - Our standard synthetic benchmark suite, updated to version 2005. While it doesn't provide real-world information, it does give us a base for the rest of the tests.

SYSMark 2004 Office and Content Creation - A scripted benchmark using real-world applications. Like the SiSoft tests, higher numbers are better.

PiFast - A good indicator of CPU/Motherboard performance is PiFast version 4.2, by Xavier Gourdon. We used a computation of 10000000 digits of Pi, Chudnovsky method, 1024 K FFT, and no disk memory. Note that lower scores are better, and times are in seconds.

TMPGEnc 2.521 - We used an Animatrix file, titled The Second Renaissance Part 1, and a WAV created from VirtualDub. The movie was then converted it into a DVD compliant MPEG-2 file with a bitrate of 5000. Times are in minutes:seconds, and lower is better.

CDex Audio Conversion Wav to MP3 - CDex was used to convert a 414MB Wav file to a 320kbs MP3. Times are in minutes:seconds, and lower is better.

Doom 3, Far Cry, Unreal Tournament 2004 @ 640x480, LQ Settings - While higher resolutions tax the video card, lower resolutions rely on CPU and subsystem speed. These results are real-world, and higher scores are better. Bench'emAll was used to collect numbers from Far Cry and UT2004.

SiSoft Sandra 2005 CPU

SiSoft Sandra 2005 MMX

SiSoft Sandra 2005 Memory

All three boards perform closely with one another, though the DDR equipped PT894 trails the pack. One area the nForce 4 SLI clearly dominates is the memory benchmark. We had to redo the tests a few more times to make sure we weren't imagining it, but the nForce 4 SLI has a clear advantage here. The main reason of is pretty much the board's ability to run at tighter timings than we've seen in the 915/925 chipsets. For our testing, the values were as such:

T(CAS) - 3
T(RCD) - 2
T(RP) - 2
T(RAS) - 7
T(RC) - 2
Addressing Mode - 1 clock

Very impressive and puts some distance in the memory tests. Some credit should go to the memory controller itself though, as timings alone don't account for the differences in performance.

SYSMark 2004 Office and Content Creation

Bit of a flip flop here as the nForce 4 SLI wins the Office Productivity tests by a small margin and places second in the Internet Content Creation test by a bit. The differences are within our margin of error though, so it's safe to say that the performance is on par with the 925XE in terms of applications.

NEXT