FPSLabs : A Brief Look at CrossFireX Scaling

Performance numbers with a super early CrossFireX evaluation driver. Not to be considered representative of final performance!

A couple of weeks ago we reviewed AMD’s new Radeon HD 3870 X2. Our conclusion, in pretty much perfect agreement with the rest of the review sites that published a week later, was that the Radeon HD 3870 X2 is the new fastest 3D graphics card in the world. We cautioned, however, that NVIDIA’s upcoming GeForce 9800GX2 would likely be faster in most tests. With recent news regarding the delayed launch date of the 9800GX2, it looks like AMD has about one month left of complete dominance of the ridiculously high-end graphics market. Funnily enough, the scores we obtained with the 3870 X2 were probably not even as good as they should have been, considering we were using a slightly earlier driver set than the one most other review sites used. Well, now there are a couple of more recent driver versions, the latest of which enables two 3870 X2’s to be run in CrossFire.

This configuration, running more than two AMD GPUs in CrossFire, is actually called CrossFireX. CrossFireX does not have to be two Radeon HD 3870 X2’s, it can also be four or three of any other capable AMD graphics card. The only catch is that there is limited real estate on the motherboard, so if you are planning to put four HD2900XTs in CrossFireX with their stock coolers, you are going to run into some problems. Furthermore, the only chipset that supports four separate graphics cards in CrossFireX is AMD’s own 790FX, which sports 4 PCI-Express x16 slots. What makes the Radeon HD 3870X2 so attractive for this configuration is that it only requires the use of two PCI-Express slots. Not only does this make it more ideal from a space conservation standpoint, but it also enables CrossFireX on several other more popular chipsets (Intel X38, X48, P35, 975, 965; ATI Xpress 3200).

[ArticleAdvert]

The original launch date for AMD’s Radeon HD 3870 X2 was to be January23rd. AMD pushed the launch back to the 28th at the last minute, supposedly because the card did not scale well in several games, including Crysis. Following this logic, AMD is also delaying the official launch of CrossFireX because they have not been able to get the performance up to where they would like. Currently, AMD is saying that CrossFireX will launch in late Q1, meaning most likely sometime in March. The first evaluation version of the CrossFireX-enabling driver was released on Wednesday, the 30th of January, and somehow beyond all expectations, it made its way to us.

Please keep in mind that this article in no way should be considered an indication of final CrossFireX performance. As far as we know, the driver being used here is the very first to support CrossFireX, so over the course of the next month we fully expect scaling and performance in general to increase quite a bit with subsequent releases. The numbers presented in this article should only be interpreted as “a brief look at the state of scaling performance generated by AMD’s current beta CrossFireX driver”. Again, the numbers presented in this article are tentative -by no means are they or should they be interpreted as directly indicative of the performance that CrossFireX will deliver upon its official release. AMD/ATI has been able to produce huge performance gains through incremental driver releases in the past (Catalyst 5.10 to 5.11 comes to mind), and we strongly suspect they will be able to do the same thing here.

We’ve already been over the features and specifications of this card in our review two weeks ago. If you are unfamiliar with this information, we suggest you check it out before reading too far into the rest of this article.
{mospagebreak heading=Introduction&title=Test Setup and Methods}
Test Setup and Methods

Please note that this hardware setup is very much different from what we usually test video cards with. The reason for this is that we had to borrow a CrossFire capable system, as our normal testbed utilizes NVIDIA’s nForce 680i SLI chipset. Pretty much all components are dissimilar between the two systems, so under no circumstances should conclusions be drawn relating scores obtained in the two articles. It is only appropriate to draw conclusions from each set of data independently.

Hardware Setup

[ArticleAdvert]

Case: Cooler Master Stacker 830
Power Supply: Cooler Master Real Power Pro 1000
Motherboard: MSI X48 Platinum
Processor: Intel Core 2 Extreme QX9770 @ 4.2GHz (400×10.5)
Hard Drive: 2x Western Digital Raptor 150 10,000RPM
Video: 2x MSI Radeon HD 3870 X2 (R3870X2-T2D1G-OC 850/1800)
Memory: 4096MB (4×1024MB) Kingston HyperX PC3 14400
Optical Drive: ASUS DRW-2014L1
Cooling: Cooler Master Hyper 212

Software Setup

Operating System: Microsoft Windows Vista Home Premium 64-bit
Video Driver: AMD 8.451.2 Quad Vista Evaluation Software

Our methodology for testing video cards typically requires time quantified in weeks. For the testing of this card, we had mere hours. For this reason, we were only able to run three of our tests in accordance with our testing methodology. There is a slim chance we will be able to run a few supplementary benchmarks over the duration of this week, the results of which will be reported in a subsequent article. The plan right now is to write an update each time AMD releases a revised evaluation driver, but hardware limitations will likely be an obstacle.

All tests were conducted at least three times, with the average score being that reported in the graphs. For Futuremark’s 3DMark06, our only synthetic test, settings were left on default out-of-the-box values. In our game tests, built-in benchmarks were used for F.E.A.R., and Call of Juarez. The benchmarks presented in this review are Vista exclusive, as we were unable to secure XP drivers in the short time we had with the card (pretty sure they do not exist).

Vsync was turned off for all tests. If there were any special kind of graphical quality settings used in any game, they will be noted in the section for that game or a screenshot of the settings will be provided.
{mospagebreak title=3DMark06}
Tests

3DMark06

Since Futuremark’s 3DMark Vantage is not yet out (though the first screenshots have recently been made available), 3DMark06 remains the go to utility for a quick benchmark. The numbers it produces aren’t necessarily indicative of real-world gameplay or gameplay in general, and for that reason we really dislike using it. However, as long as the rest of the world uses it to evaluate hardware performance, we will too. This test was run first on a single Radeon HD 3870 X2 multiple times and then again when the second card was installed (and then back to single card to take the screenshot we forgot to take).

{multithumb popup_type=none enable_thumbs=0} {multithumb popup_type=default enable_thumbs=default}

[ArticleAdvert]

Here you can see that scaling in 3DMark06 at the default 1280×1024 resolution would appear to be unimpressive. However, scaling in 3DMark06 is typically quite bad, considering the benchmark’s often overpowering CPU dependence. If you break down the numbers a bit more, you will see that most of the improvement between configurations comes from the HDR/SM3.0 Score. This is not very surprising, as increasing graphics memory and computing power lends itself to increased performance in more recent graphic instruction sets. Scaling in 3DMark06 looks to be at best 16%, and averages somewhere around 15%. This might seem low, but as we mentioned before, 3DMark06 does not usually scale very well, as it seems to depend more on the CPU when at default settings. For this reason, it is tough to say whether or not there is room for improvement with CrossFireX scaling in 3DMark06. We were unable to run the test at higher resolutions and quality settings due to time restrictions, but more pertinent information could probably be obtained by doing so.
{mospagebreak title=Call of Juarez}
Call of Juarez

Call of Juarez was another game we had on hand and could be benchmarked fairly quickly, which was our primary concern while testing these cards. Whereas many games offer pretty poor built-in benchmarks, the Call of Juarez DirectX 10 test incorporates a lot of different gameplay elements that make it a pretty good indication of in-game performance. The game is also ideal for testing high-end graphics cards, as in DirectX 10 in looks quite pretty and brings even the best graphics subsystems to their knees at higher resolutions. Call of Juarez certainly isn’t the best-looking game around anymore thanks to Unreal Engine 3, updates to the Source Engine, and Crysis, but it will remain in our benchmarking suite until it proves to be useless.

{multithumb popup_type=none enable_thumbs=0} {multithumb popup_type=default enable_thumbs=default}

[ArticleAdvert]

Here you can see that scaling in Call of Juarez is quite good. We ran the CoJ test five times for both graphics configurations and all tests were within one tenth of the first value. Scaling ended up being very solid at 50%. Video cards from ATI have always scaled much better than their NVIDIA counterparts in Call of Juarez, and a quick look back at our 8800GT SLI review will reinforce this point. That said, there is probably slight room for improvement here, though AMD shouldn’t be in a hurry since they are already quite far ahead in this game.
{mospagebreak title=F.E.A.R.}
F.E.A.R.

By now F.E.A.R. is a pretty old game, and many might question why we would even incorporate it into our benchmark suite. Well, despite several drawbacks, F.E.A.R. was actually played competitively for a while and for good reason: it is damn fun to play in multiplayer. The built-in benchmark is probably a bit too slow to represent real gameplay, but due to the incredible range of explosions and shots and particles that can be flying around while playing the game, it is definitely the most reliable method. F.E.A.R. has also always exhibited incredible scaling performance in multi-GPU configurations for graphics cards from both AMD and NVIDIA, so it will be interesting to see what four GPUs can do against the game.

{multithumb popup_type=none enable_thumbs=0} {multithumb popup_type=default enable_thumbs=default}

[ArticleAdvert]

These are some pretty impressive results for the two Radeon HD 3870 X2’s in CrossFireX. Scaling is very close to 80% in F.E.A.R., which is pretty much in-line with what we have come to expect from multi-GPU configurations in this particular test. It would seem as though increasing from two to four GPUs boosts frames just as much going from one to two. There is very little room for improvement here from AMD, as the theoretical max is 100% and as-is they are just 20% off that mark.
{mospagebreak title=Conclusion}
Final Thoughts and Conclusions

In case we haven’t made ourselves absolutely clear by now, the numbers presented in this article were obtained using the earliest evaluation version of CrossFireX-enabling drivers. In no way are they directly indicative of the performance that will be available when the final CrossFireX driver is launched to the public (probably sometime in March). Also, before comparing these numbers to those you might see on other sites, please make sure you check out the specifications of the test system we used and consider applicable differences and their potential affect on the scores.

{multithumb popup_type=none enable_thumbs=0} {multithumb popup_type=default enable_thumbs=default}

[ArticleAdvert]

From what we can tell so far, it looks like AMD’s CrossFireX technology is already in pretty good shape with this first driver release. Back during the Radeon 8500 driver fiasco, ATI lost a lot of credibility when it came to the driver game. More recently however, the regular updates, general stability and feature set of AMD/ATI drivers has more than brought them back up to the expected level of quality. AMD seems to have adopted a fairly long development and testing period for driver releases, and makes it a point to launch a new version monthly (which is more than can be said for their counterpart).

With numbers like this, it will be very interesting to see what kind of additional gains AMD can squeeze out with further versions of the CrossFireX driver. It is also very possible that the numbers you’ve just seen do not paint the full picture of where this driver is currently. Hopefully this possibility can be ruled out or confirmed with further testing in various other games. As mentioned in the testing methodology we are also going to make an attempt to provide you with updates each time AMD releases a new evaluation driver if we are fortunate enough to have it land in our laps again.

Popularity: 53% [?]

Support FPSLabs! Share and Enjoy: