Tech-Hounds.com

Crossfire: ATI's Answer to SLI - Part 2

Both Crossfire and SLI took some news time the past weeks with NVIDIA and ATI making the annoucement of supporting Havok FX physics. NVIDIA was the first to announce their support, back at E3 and ATI has followed suit in Computex. In general we think this is a good thing, particularly with SLI. This means if you don't experience a performance increase with a Havok FX enabled game in SLI, you can either use SLI to have SLI AA or physics. At the moment, ATI certainly have the better offering here, because you can pair different cards - say a X1900 with a X1600 - one to do the graphics and one for the physics. God knows you don't want to use such a combination for graphics.

However, what is a dissapointment is the two manufacturer's insistence to enable physics acceleration only on their own platforms - ATI with Crossfire and NVIDIA with SLI. Those who use Intel processors with Intel dual PCI Express x16 slots may get them (or may not). So, users with AMD processors most likely have to make a choice, it's either Crossfire or SLI. The only platform independent solution at this time is Ageia's PhysX and that hasn't impress many people, users and reviewers alike.

We all can guess why ATI and NVIDIA choose to only support physics accelaration with their own platforms, but it is dissapointing nonetheless. It's probably as dissapointing as having those dual GeForce cards, apparently you still have to use NForce 4 and 5 SLI series to use these cards at all. Thankfully, since both manufacturers support the same software (Havok), there will likely be no such issues with Havok FX enabled games (once they arrive).

Since there are no Havok FX enabled games that support physics acceleration in the market at this time, we'll have to evaluate each manufacturers' solution by what's possible right now - graphics performance and quality.

After seeing what Crossfire has to offer with the Radeon X1600XTs, we really want to know just how far you can go. Thankfully, we've managed to snag a Radeon X1900 Crossfire Edition and a Radeon X1900XTX for some Crossfire testing. Like we promised last week (actually that's two weeks ago - ed), we will not only be looking at what a pair of Radeon X1900 can offer in Crossfire mode in terms of performance, but quality as well.

In case you don't know, if you want to use two Radeon X1900 in a Crossfire setup, you'll have to put up with the Crossfire cable sticking out of the case. These high end cards still use the old version of Crossfire. Obviously, this has aroused some complaint from many reviewers and users - us included. After all, the Radeon X1800GT (and quite possibly the Radeon X1900GT) can be used in Crossfire mode without cables and a master card. Well, only ATI knows the answer to that, however we think there's a technical explanation for this. With dongless (and masterless) Crossfire, all card to card communication have to go through the PCI Express bus. The bandwdith used to send the second image (in AFR mode) to the first card can be very high (think 1600 x 1200 at more than 60 fps). That's true even if the two cards use full x16 slots. There's also the additional latency to account for. Since there's no bridge or internal connector like those present in GeForce 6 and 7 series of cards, ATI have to make do with an external cable. Since it's an external cable, that means its very susceptible to improper installation or worse, faulty parts and manhandling.

This article took longer than expected because of some problems we have with some of our testing procedures. Mind you, these are not technical problems concerning either the Radeon cards or the Crossfire motherboard. Unlike the two Radeon X1600XTs, you still need to use a Crossfire cable for the Radeon X1900. Like most external cables / connectors, improper installation can cause some distortion. When we did our preliminary testing with the two Radeon X1900s, there's a faded, vertical white line on the left side of the screen. This line is only noticeable at bootup (or desktop resolution changes). It's not necessarily distoring the screen, it's just simply annoying that it's present. After all, a Crossfire setup like this cost quite a lot of money and we certainly don't appreciate something like this. We also notice some flickering issues with some games. Once we make sure the cable is properly installed (the second time), most of these issues went away (or at least become less noticeable).

However, the most important problem to mention is image quality comparison, which we will elaborate.

Image Quality Comparison

Crossfire's Super AA in many ways is similar to SLI's SLI AA. Both cards render the same image with anti aliasing enabled, but the samples are taken from different parts. However, ATI is quick to point out that their Super AA sample patterns are not simply 4x or 6x sample patterns with slightly difference placement. That's because Super AA sample patterns make use of ATI's programmable sample hardware to retrieve samples for anti aliasing. So, theoretically, ATI has much better anti aliasing than NVIDIA since they can customize the sample pattern for Super AA.

However, what's more ingenious is the extra modes ATI made available with Crossfire - Super AA 10x and 14x. In addition to using eight and twelve samples for anti aliasing the image you see on the screen, ATI have also found a way to provide super sampling anti aliasing as well. This works by taking the first sample from the first card and the second from the second card. So, Super AA 10x is in someways pretty similar to what you can achieve with 8xS with NVIDIA's hardware (8 MSAA samples and 2 SSAA samples). 

Let's have a glimpse of what Super AA has to offer. First, here are sample patterns for 4x and 6x AA.

 

And here are the Super AA sample patterns (8x, 10x, 12x, 14x)

     

If you were looking at these images closely, you'll no doubt notice that only half the number of samples are displayed in the Super AA 8x and 12x images. This is a quirk with Super AA (or at least the current drivers) that makes it impossible for us to capture what Super AA has to offer in terms of image quality. The screenshot can only capture output from one of the cards, so only half the sample is captured. Trying to capture shots or sequences from games also yield the same results. In light of this, we will not be doing any image comparison test using games for this article. Another quirk of Super AA is that it will only work with fullscreen games and applications. That means you only get 4x AA  if the 3D application or game you're running is in windowed mode.

So, do we have to settle for 'theoretically better' argument? Not quite - look again at the images. You'll notice that the lines near the edges of the screen with Super AA 10x and 14x have less aliasing than 4x or 6x AA. We can assure you that even Super AA 8x and 12x have the same aliasing as 4x and 6x (though it's slightly less but still more noticeable than 10x and 14x). We know. It's hard to believe what Super AA 10x and 14x has to offer unless you're looking at them in real life. But it looks like this will have to do for now.

Performance

AA and AF settings are applied in game, whenever possible. We had to apply AA and AF from the driver panel for Call of Duty and Richard Burns Rally. In Quake 4, we enable both AA and AF from the console, with the command 'r_multisamples' set to 4 and 'image_anisotropy' set to 16. V sync was disabled both from inside the game and on the driver panel. AA and AF is enabled from the settings menu for F.E.A.R, Serious Sam II and Splinter Cell: Chaos Theory.

Call of Duty and Richard Burns Rally was configured to the highest possible detail. F.E.A.R settings are set to our usual test settings (all settings set to maximum / on except for soft shadows). Quake 4 was set to 'High Quality' while Serious Sam II was tested with our usual standard settings (maximum settings but with custom resolution, anti aliasing and anisotropic filtering settings). The same also applies to Splinter Cell Chaos Theory - the game was benchmarked to use its shader model 3.0 and everything turned on, except for HDR which is only turned on for HDR performance tests.

We ran our usual tests under two different configurations - one with a single Radeon X1900XTX and another with a Radeon X1900 Crossfire Edition master card and the Radeon X1900XTX as a slave card. Catalyst AI was enabled so it will apply the best Crossfire mode (alternate frame rendering)

We'd like to thank ASUS for supplying the ASUS A8R32-MVP Deluxe motherboard, the Radeon X1900XTX and CrossFire Master card as well as Tagan for supplying the additional power supply for this article.

Our test setup
AMD Athlon 64 3500+ socket 939
2 x 1024 MB Kingston KVR 3-3-3 PC3200 DDR-SDRAM
ASUS A8R32-MVP Deluxe Radeon Xpress 3200 Crossfire motherboard
ASUS Radeon X1900 Crossfire Edition DDR3 512 MB graphics card
ASUS Radeon X1900XTX DDR3 512 MB graphics card
Maxtor DiamondMaxPlus9 80 GB Serial ATA 8 MB buffer
ASUS E-616 DVD-ROM
Tagan TG530-U15 530 watts ATX/BTX power supply

Windows XP Professional with Service Pack 2 installed
ATI Catalyst 6.5 reference driver
Realtek High Definition Audio 5.10.00.5202 driver.
DirectX 9.0c

Performance

The results:

The graphs are pretty self explanatory, but in case you can't see the text, they're arranged in groups of three colors: green for minimum fps, blue for average fps and red for maximum fps. The first group represent test results under default settings, and the second represent results with 4xAA and 16xAF enabled.  For Serious Sam II HDR tests, the second groups represent results with HDR, AA  and 16x AF. These results are in frame rates per second.


With such a high performing card like the Radeon X1900, we won't see dramatic increases like we did with two low end cards. Two Radeon X1900 in a Crossfire setup is not going to offer you higher frame rates than a single card in a system limited setup such as this. We only begin to see a difference at 1280 x 1024, where the single Radeon X1900XTX is trailing behind by 10 fps or so. That's not really significant with the results we're seeing here - about 5 percent slower. The trend continues at higher resolution, and here we see an even bigger gap between the two. It's interesting to note that with higher resolutions, even with Crossfire, we're getting slightly lower frame rates. So, maybe we're not that system limited after all.

Since Call of Duty is an outdated game, this is not unexpected. It's very likely we will see a similar trend with most, if not all, pre DirectX 9 games. In these games, having two Radeon X1900 in Crossfire mode allows you to have slightly higher frame rates in situation where the game is texture / pixel shader limited (usually by enabling AA / AF). From the results, we could see that the added pixel processing power of a second card will let offer less penalty when AA and AF are used in higher resolutions. For Call of Duty, instead of losing performance at about 15 - 25 percent, you're only losing 5 - 6 percent.

It's really not a great argument for going Crossfire. But of course, the extra processing power from the second card allows you to turn on graphical features that was prohibitively expensive with a single card, like adaptive anti aliasing. The next game we benchmark is particularly interesting for that reason alone. Let's take a look.


We had to rename the executable file so that Catalyst AI will force Crossfire AFR mode with this game. Looks like even a single Radeon X1900 is enough - we're not seeing any difference until we hit 1600 x 1200 with AA and AF. We're definitely system limited here. A difference of 6 to 15 fps with 4x and 6x anti aliasing is not that strong a reason to get two Radeon X1900 in a Crossfire setup. However, we're not really interested with default and AA / AF results with this game.


Turn on adaptive anti aliasing with 'Quality' option and now we can see a stronger case for a Crossfire setup. Right off the bat, at 1024 x 768 we can already see quite a huge gain with a Crossfire setup - roughly 15 percent in 4x AA and 50 percent in 6x AA. At higher resolutions, the difference grew to about 70 - 80 percent at 1600 x 1200.. You only have enough processing power to run 4x adaptive anti aliasing at 1280 x 1024 with a single card, but put two Radeon X1900 together, you can push it for 6x AA at the same resolution. You might even be able to get away with 1600 x 1200 if you can stand minimum frame rates near the 30 fps mark.

There's no doubt, when Crossfire mode is enabled, you get a tremendous boost of pixel processing power. Using two Radeon X1900 cards will only make sense if you play at the highest resolution, all details maxed up with AA and AF for older games. You might even get away with enabling adaptive anti aliasing, a feature that's a tremendous burden, even for a high end card such as the Radeon X1900XTX.

Looking back to our last test, we can see a pair of Radeon X1900 behaving quite differently compared to a pair of Radeon X1600 in Crossfire mode. We didn't see any increase under default settings or even AA / AF in most cases. However, the reason is actually quite simple - we're system limited with these games. If we had used a faster processor, we will likely see the two Radeon X1900 will also offer frame rate increases across the board. So, let's test this hypothesis with games that are not system limited.


Undoubtedly the heaviest game around before The Elder Scrolls IV: Oblivion came out, F.E.A.R is a game where you expect Crossfire to shine. Even with default settings at 1024 x 768, we can already see the two Radeon X1900 offer a slightly higher frame rate overall - a 10 fps difference in average fps. That's about 10 percent faster. With AA and AF enabled, the difference grew to 20 fps or about 25 percent faster. Crossfire,  might even let you get away with 6x AA and AF or up the resolution to 1280 x 960, something a single Radeon X1900 can't offer. But even a Crossfire setup lacks the huge processing power to push this game at the highest setting at 1600 x 1200, unless you're planning to play without AA and AF.

Looking back at the single Radeon X1900XTX results with default settings at 1024 x 768, they're actually pretty close to what we're getting with Crossfire at 1600 x 1200. Making an estimated guess, we calculated that the two Radeon X1900 cards in Crossfire mode roughly offer about 70 percent increase in pixel processing power compared to the single Radeon X1900XTX.. Since we're not system limited with F.E.A.R (after all the Crossfire results are faster than a single card), there must be another factor limiting Crossfire performance. We think it's actually vertex shader limited, since these shaders are not load balanced between the two cards.

One concern with this game is the slightly lower minimum frame rates we're getting. This could be caused by the newer drivers (Catalyst 6.5) or the use of a different board. We usually saw a higher minimum fps from the Radeon X1900XTX, about 5 - 10 fps faster. We'll have to look further into this, but for now, let's move on to the next game in our benchmark suite.