While doing some more game testing for an upcoming review, I came upon an issue that seemed quite confusing.  While playing Prey, the new Doom 3 engine game from HumanHead Software and 3D Realms, I noticed that in at least one case, the X1950 XTX card I was using scored the same when set at 2xAA and at 4xAA.  Looking back at my review of the X1950 XTX card, specifically the Prey gaming section, I did not test the non-AA or lower-AA settings; hence why something like this wouldn’t have showed up before hand. 

So, the 2xAA and 4xAA scores were the same in one game; obviously some kind of bug in the driver.  What was even more odd is that the X1950 XTX CrossFire Edition card did NOT replicate this behavior.  This tripped another warning in my head and I checked to find out that this occurs in all the Doom 3 engine games I have here: Doom 3, Quake 4 and Prey.  No other titles exhibited this type of behavior and it didn’t occur at any resolution other than 2560×1600. 

After doing some more testing, I decided to graph the results:

ATI X1950 Issue: CrossFire and Standard Card Render Differently? - Graphics Cards 5

ATI X1950 Issue: CrossFire and Standard Card Render Differently? - Graphics Cards 6

So, here we see the theory unfold; the standard X1950 XTX card falls back into the 2xAA mode even when 4xAA is requested and set in the game.  The CrossFire card though, seems to follow the rules correctly, as we see a noticeable performance drop moving from 2xAA up to 4xAA, as you would expect.  That’s a difference of about 15% in real-world game play results, no time demos here.  Unless the CrossFire card is doing some magic we weren’t told about…?

To check myself, I took some screen shots from the same spot in the game: one with the standard X1950 XTX set at 4xAA and one with the CrossFire card set at 4xAA.

ATI X1950 Issue: CrossFire and Standard Card Render Differently? - Graphics Cards 7

Click to Enlarge

ATI X1950 Issue: CrossFire and Standard Card Render Differently? - Graphics Cards 8

Click to Enlarge

X1950 XTX Standard – 2560×1600 – 4xAA

X1950 XTX CrossFire- 2560×1600 – 4xAA

You can see in these screen shots from our run through of Prey that the X1950 XTX standard is using a lesser-quality AA even though it is set at identical 4xAA in the options. 

This same affect can be seen in Quake 4 as well as Doom 3; though I didn’t have time to do the real-world gaming results I would have liked, here are some quick time demo results from the in-game benchmarks that come in the retail versions.

Quake 4 – demo3 – 2560×1600

2xAA

4xAA

X1950 XTX Standard

76.2

76.1

X1950 XTX CrossFire

76.1

54.5

Doom 3 – demo1 – 2560×1600

2xAA

4xAA

X1950 XTX Standard

57.1

57.2

X1950 XTX CrossFire

57.1

40.3

So while these timedemo runs don’t represent real game play, they are incriminating none the less.  Quake 4 shows a 41% performance hit when in the “real” 4xAA mode on the CrossFire card that the standard card doesn’t get hit with.  In Doom 3, the same is happening at a 41% performance delta as well. 

Whether or not this was an accidental bug or something else, there are probably numerous reviews online about the X1950 XTX cards that need to be re-evaluated, my own included.  Any readers or reviewers that were overly impressed by the X1950 XTX’s performance in OpenGL applications (possibly just limited to the Doom 3 engine) need to run these kinds of tests themselves to see if their results are tainted by this error.  Hopefully ATI will address this issue quickly and in the right manor!

Jump into this thread in our forums to discuss this issue: can any CrossFire users replicate it?  Any other titles (besdies these OpenGL/Doom 3 ones) showing this behavior?

Update 11/02/06 – To answer some questions that have come up, the driver version I used to get the numbers above was Catalyst 6.10, default settings in the control panel.  I have tested 6.9 and 6.8 and saw the same issue on these three games using the original X1900 XTX and X1900 CrossFire Edition cards as well (since 6.10 was the first WHQL driver to add X1950 XTX support, only it was used on that card).  The fall back is not necessarily a bad thing, if ATI had told us or indicated to us that this was happening; they did NOT do so though.  Add to that the fact that their CrossFire edition card DOES NOT EXHIBIT THIS BEHAVIOR and you might finally see the reason for us bringing it up.