Multi-GPU World Tour 2006 Part 5: Most Played Titles

Author: Geordan Hankinson
Editor: Howard Ha
Publish Date: Saturday, July 22nd, 2006
Originally Published on Neoseeker (http://www.neoseeker.com)
Article Link: http://www.neoseeker.com/Articles/Hardware/Reviews/multigpumostplayed/
Copyright Neo Era Media, Inc. - please do not redistribute or use for commercial purposes.

Introduction

Many weeks ago Neoseeker started work in a one of a kind collaborative effort with 8 other well known hardware sites to create a Mult-GPU series of articles that would span across 8 sites, cover over 60 benchmarks, and put 3 ATI CrossFire and 3 NVIDIA SLI systems under the microscope.  Last week the article series was launched and 4 articles have been published since.  This is now part 5 of the Multi-GPU World Tour and we'll be looking at the "Most Played Game Titles".

By this point if you've read the previous four articles from PenStarSystems, Bit-Tech, Bjorn3d, and HARDiNFO, you should have a fairly good impression of what we're looking to accomplish with this multi-site benchmarking article. As far as actual numbers go, we've seen a number of motherboard benchmarks from Bjorn3d, and on Wednesday HARDiNFO posted the first of the Multi-GPU video card setup results. HARDiNFO was responsible for covering the most frequently used benchmarks in hardware testing and their massively comprehensive coverage gives a good idea of the trends seen by many websites in their testing.

For those who are joining us without having read the previous 4 parts of this multi-site, multi-GPU article, I strongly encourage you to go back and check out those articles linked below which outline the hardware involved, the motherboards, and the initial slew of benchmark results.  This series is intended to read almost like multiple pages of a massive article, so please do check out the other installments to get the whole picture.  Here's an "index" of already published and upcoming articles:


Our contribution has us looking at a rough cross section of a few of the most played multiplayer games on both XFire and Gamespy, though we do cover a couple more recent high profile titles such as Oblivion.  Not surprisingly, the titles at the top of XFire and Gamespy's list are not at the very bleeding edge of graphics - they require a certain amount of horsepower certainly, but these games are played by a wide range of users who have a variety of system capabilities.  This puts our particular tests in a less exciting light as far as sheer horsepower tests are concerned, but it does make these results highly relevant and representative of what a large gaming audience can expect from Multi-GPU setups.  Here's the rundown of titles we cover in this article:

Oblivion
Sin Episodes
Counter Strike Source
Wolfenstein: Enemy Territory
Call of Duty
Battlefield 2
Guild Wars: Factions
World of Warcraft
Warcraft III

All of the Editors behind this article agreed that it would be best to go about the specifics of each test as saw fit. Because there are already hundreds of published benchmarks for the different cards being used in this testing, we at Neoseeker did not deem it important to feature standard settings across the three platforms.  As a result, each segment (low, mid and high end) feature varying differences in terms of resolution, filtering and detail options, that are tailored to that platform. The high and mid-end setups that we all agreed on are very close in terms of performance and as a result we decided on identical settings  across those two platforms because the results were so closely packed. Our low-end setups however were not capable of the high filtering and resolution combinations used in some instances so there is a marked difference in what we use in some of those cases.

Unfortunately, Battlefield 2, which is one of the games that we were tasked with benchmarking did not make it to the final cut of the article.  We ran into some problems with the game a third of the way through testing and were unable to rectify the situation in time for this article despite repeated re-installations and numerous other attempted tweaks. We will be sure to update our benchmarks once we sort out the issue.

Read on for a look at our hardware setup.

Test Setup

For the most part, all sites involved responsible for benchmarking are using identical hardware setups. There are some instances where  we may be using completely different motherboards, however these are always based off of the same platform leaving little room for performance deviation.

We will be going over the specifics of each benchmark and the settings used in each case on the respective pages of each benchmark, and be sure to note that we are not using the same settings across all platforms in some circumstances (Oblivion being the primary example) so take note of any text!  Many of the games tested do not have any built in timedemo or benchmarking capability, so we turned to the use of the always trusty FRAPS application we're all so fond of.

As far as our driver settings go, we have image quality set to the highest possible setting for both ATI and NVIDIA cards. Any optimization options are disabled as are Transparency and Adaptive Anti Aliasing.

One important note that has been discussed already but is worth mentioning again is in regards to the cableless Crossfire being enabled on X1900 GT cards.  ATI made a driver available which allows for two of these cards to be used in a masterless setup on Crossfire motherboards. While we haven't looked into the specifics of using two X1900 GT's in a cableless Crossfire setup, be sure to check out our testing with two X1800 GTO's in an identical scenario.

ATI managed to get a pre-release copy of Catalyst 6.7 to all the editors today, but we simply did not have the time to incorporate results with the new drivers in place.

Our NVIDIA systems are using driver version 91.31 while our ATI systems are using driver version 6.6 with the exception of the mid end ATI setup which uses the Catalyst hotfix driver available here.

High End Systems

7900 GTX SLI

AMD FX-62 AM2
2 GB DDR2-800 4:4:4:8
ASUS M2N32-SLI

X1900 XT CrossFire

AMD FX-62 AM2
2 GB DDR2-800 4:4:4:8
ECSKA3-MVP

Mid End Systems

7950 GX2

AMD X2 4400+ 939
2GB DDR PC3200 2:3:2:7
Abit 
AN8 32X

X1900 GT
AMD X2 4400+ 939
2GB DDR PC3200 2:3:2:7
Abit 
AT8 32X

Low End Systems

7600 GS

AMD X2 3800+ 939
1 GB DDR PC3200 2.5:3:3:7
DFI LanParty UT NF4 SLI-DR

X1600 Pro

AMD X2 3800+ 939
1 GB DDR PC3200 2.5:3:3:7
Abit AT8

Guild Wars: Factions

Our chart setup for this article fails to note the filtering settings which is unfortunate, but we'll note what settings are used in these blurbs and note when settings change across platforms.. In the case of all Guild Wars across the top two platforms, all settings are at their highest, although we use 8x AA and 16x AF enabled from the Control Panel. Our low end setup uses only 4x as opposed to 8 as well as lower resolutions.



Oblivion

Oblivion is arguably the most graphically intense of all the titles we tested.  You'll note that as you browse through the results.  All details in Oblivion are set to maximum for this test (as well as for the mid end platforms) though HDR is disabled. We're using the ingame 4x AA setting with 16x AF enabled from the control panel. Our mid end testing uses identical settings and resolutions as the high end setup.  For testing we used Fraps to records framerates in a specific path.  Our Fraps path moves from the Cloud Ruler Temple down into the valley towards Bruma.



Sin Episodes

All settings are at maximum with 4x AA and 16x AF enabled in-game. Setting 8x AA from the control panel instilled far too much of a performance hit to be practical.



Counter Strike Source

Probably one of the most played games by several of our friends and editors.  Identical settings used here as in Sin though do note that Valve's HDR effects have been disabled.

Warcraft III

Warcraft III is easily one of the oldest titles still being played with such dedication by fans, this is a true testament to its staying power.  Unfortunately its age also means that Multi-GPU results are not going to look as exciting.  All settings are at maximum here with 8x AA and 16x AF enabled from the control panel. Most of the numbers we received in this game at the lower resolutions are not a product of being CPU bound, but are caused from the 64 FPS cap that we were unable to find a work around for. Our sixty second FRAPS sequence generally keeps alot of characters on screen at a time and does reveal minor performance differences at higher resolution.



World of Warcraft

All in game settings were at their highest although 'glow' was was disabled to allow for AA support. We ran our tests with 8x AA and 16x AF enabled from the control panel. Because of the dynamic nature of the world with player characters moving around, testing in most areas is impractical.  Our test was conducted with Fraps on a gryphon flight from Stormwind to Westfall over and through a forested area.  This is a repeatable test that gave us stable enough results to report on.  Note also that World of Warcraft also has a 60-64FPS cap that adversely affects results.



Call of Duty

For Call of Duty we used a custom timedemo for testing on Carentan with 8x AA and 16x AF enabled from the control panel.



Wolfenstein: Enemy Territory

A custom timedemo on Seawall Battery was used with 8x and 16x filtering levels.



The Crossfire setup seems to have some issues with the source engine in our run of tests (though the numbers still being well above 60, this shouldn't be an issue), and Oblivion performance is down a notch as well. NVIDIA runs into an issue with Wolfenstein at 2048x1536, but that 100 FPS difference doesn't seem so large when it's impossible to detect anyhow.

Call of Duty

8x AA, 16x AF



Counter Strike Source

4x AA 16x AF



Wolfenstein Enemy Territory

8x AA, 16x AF



Guild Wars: Factions

8x AA 16x AF

Oblivion

4x AA, 16x AF



Sin Episodes

4x AA, 16x AF



Warcraft III

8x AA, 16x AF



World of Warcraft

8x AA, 16x AF



Again, the two platforms are perform similarly to eachother though NVIDIA does have an edge. With the exception of an anomaly that we couldn't seem to correct in Counter Strike Source at 2048 on the Radeon setup, and some large gaps in Guild wars and Enemy Territory at a couple of the resolutions, both run fairly neck and neck. Oblivion being the only game in our suite that seriously taxes these systems, the ~7 frames per second lead of the GX2 at 1280 and 1600 is notable.

Call of Duty

8x AA, 16x AF. Also note the lower resolutions used.



Wolfenstein Enemy Territory

8x AA, 16x AF



Guild Wars: Factions

4x AA 16x AF and lower resolution as indicated on the chart.



Oblivion

After some extended setting testing, we ended up with this configuration: Medium texture settings, no AA, Distant Land On, No Foliage Shadows, No Distant Trees, Distant Buildings on, and all sliders set to a third of the way to maximum. Also take note of the lower resolutions used.

Sin Episodes

4x AA, 16x AF. Note the lower resolutions used for both source tests.



Counter Strike Source

4x AA, 16x AF



Warcraft III

8x AA, 16x AF



World of Warcraft

4x AA, 16x AF



The two platforms keep within fairly close range of eachother though ATI manages to gain some significant ground in both Call of Duty and Oblivion. ATI's Oblivion performance in this instance is very good, leading by a wide margin.

Conclusion

While the overall conclusion is up to Guru3d to decide, the leader in most the tests we were responsible for was NVIDIA. Though not a massive delta by any means, there tends to be a relatively predictable difference between the ATI and NVIDIA numbers. This is reversed in some instances, though the majority of tests are in fact led by NVIDIA, however small that lead often works out to be in the scope of things.

Although this article series leans towards a heavy slant of comparing ATI performance versus NVIDIA, our particular portion of the series gives us an opportunity also to remark on the overall value and relevance of Mult-GPU.  We're looking at the most popular games being played, and many of these games simply don't stress the GPU enough for a very expensive setup to be of much gain. While other articles will and have covered games that put more pressure on the GPU, the only title we tested that really stresses the GPU is Oblivion.  While its well accepted now that most gamers buy new hardware to play newer games at higher detail, these results show that alot of the games that people are still playing don't require a new top end SLI or Crossfire setup.  Also of note is the fact that LCD monitors are locked at their native resolution, which in many cases works to be 1280 x 1024 and 1600x1200. With response times no longer a problem, many gamers have moved over to LCD's making this another area of consideration. With the exception of Oblivion, all of the games we tested ran at maximum detail levels just fine at 1280 x 1024 across the entire tested hardware spectrum. Increasingly, gamers looking to spend mega money on a new multi-gpu setup ought to be looking at a new, larger display (assuming they only wish to buy LCD) which supports resolutions that will truly tax the hardware, or else those same gamers really need to re-evaluate whether they need all that horsepower if they tend to focus on games like those we covered.

SLI and Crossfire have been out for a while and can be considered relatively mature technologies, but as far as issues and problems go, neither company has an entirely clean performance record during this test series. It is in the nature of a dual card setups to misbehave and output puzzling results in benchmarks and the platforms we tested here are no different. Some benchmarks (including some of our own) show that in some instances a single card solution is faster when the GPU is not the limiting factor. The extra data required to be shuffled across the bus can slow things down in some situations in comparison to a single GPU setup.  I think you'll see more of the same sort of thing as this series continues and more and more benchmarks are being put to the test.  Remember, in the following parts of the Multi-GPU World Tour the other sites will be showcasing results which represent the first time that some of those games have ever been official benchmarks in SLI and Crossfire testing of this scope.

The only major issue we ran into was the melting of our Thermaltake Toughpower 550 Watt power supply in the face of an X1900 XT Crossfire setup. Admittedly, it was my own fault for not realising that this particular supply was Crossfire certified, but only for up to two X1800 XT's. Still, this just proves the fact that high end Multi-GPU setups require solid power (though NVIDIA's power requirements are in fact less), and right now ATI Crossfire is notorious for its pickiness as far as power goes.  In fact, we recently switched our scope for PSU testing to cover Crossfire X1900 certified PSUs because we noticed the instability with X1900 setups.  With the trend moving towards heavier power draw in the next generation of GPUs, the power supply is most definitely as key a component as any other in a high end setup.

We will be looking to rectify our Battlefield 2 situation as soon as possible, and we will update our benchmarks accordingly once we figure out where the issue is.

We'd like to take this opportunity to thank the other sites in the Multi-GPU World Tour and also thank the sponsors of this series, this collaboration has been an interesting project and we're hoping that you, the reader enjoy this unprecedented article series.  Please do look for the next installment in this series from NVNews, as they will be covering the first in the series of "uncommon benchmarks", meaning... testing with games that you pretty much never see get used in hardware reviews!

Next Up: NVNews kicks off the first of the "uncommon benchmarks"

As a recap of the series, here's a complete list of the full article series "index" that was included in the introduction of this article, both released articles and upcoming installments are listed:

»Neoseeker.com

Copyright Neo Era Media, Inc., 1999-2014.
All Rights Reserved.

Please do not redistribute or use this article in whole, or in part, for commercial purposes.