Crossfire introduces a number of extra modes of FSAA in relation to a single board, with 8x, 10, 12x and 14x modes. These higher quality modes operate by having each of the board render either 4 or 6 samples, each with different sampling positions, and then these are combined on the master card giving the higher levels of AA - more of this operation can be read here. Note that the 10x and 14x modes are rendering the same number of samples, except that the texture positions are offset between each of the boards, giving an element of texture anti-aliasing as well; as the performance of these modes are the same we'll look at 8x (4x FSAA x2) and 14x (6x FSAA x2) Super AA.
Far Cry - 4 samples per board + 8xAF  
|
307200 |
480000 |
786432 |
1310720 |
1920000 |
X1900 XTX |
30 |
47.3 |
76.8 |
125.6 |
178.6 |
X1900 XF |
29.6 |
46.5 |
76.3 |
124.9 |
169.2 |
X1800 XT |
30 |
46.9 |
76.5 |
125.4 |
174.5 |
X1800 XF |
29.8 |
46.4 |
76.1 |
123.9 |
160.5 |
pixels x fps - Mpixels per second
Crossfire 8x Super AA performance difference  
|
640x480 |
800x600 |
1024x768 |
1280x1024 |
1600x1200 |
X1900 XTX |
-1.50% |
-1.80% |
-0.70% |
-0.50% |
-5.30% |
X1800 XT |
-0.50% |
-1.10% |
-0.50% |
-1.30% |
-8.0% |
percentage
With our Far Cry test the X1900 Crossfire is rendering 8x FSAA (4x per board) at only 5% slower than a single board is rendering 4x FSAA, with the average rendering performance at just a few FPS shy of 90. Bear in mind that the Crossfire Edition board is slower than an XTX board, hence the overall Crossfire performance shouldn't be double that of a single XTX.
Far Cry - 6 samples per board + 16xAF  
|
307200 |
480000 |
786432 |
1310720 |
1920000 |
X1900 XTX |
29.9 |
46.9 |
76.8 |
124.6 |
169.5 |
X1900 XF |
29.5 |
46.3 |
76 |
124.5 |
166.5 |
X1800 XT |
30 |
46.9 |
76.5 |
125.4 |
174.5 |
X1800 XF |
29.6 |
46.5 |
76.2 |
121 |
149.2 |
pixels x fps - Mpixels per second
Crossfire 12x Super AA performance difference  
|
640x480 |
800x600 |
1024x768 |
1280x1024 |
1600x1200 |
X1900 XTX |
-1.20% |
-1.40% |
-0.90% |
-0.10% |
-1.70% |
X1800 XT |
-1.10% |
-1.0% |
-0.40% |
-3.60% |
-14.50% |
percentage
With 12x FSAA on X1900 Crossfire in this Far Cry test there is a similar situation, except the difference to a single XTX rendering 4x FSAA and the X1900 Crossfire system is even less in this scenario. Curiously, the performance difference between the X1800 XT and X1800 Crossfire increases fairly highly at 1600x1200.
Doom3 - 4 samples per board + 8xAF  
|
307200 |
480000 |
786432 |
1310720 |
1920000 |
X1900 XTX |
45.8 |
69.1 |
104.3 |
135.1 |
155.3 |
X1900 XF |
44.7 |
68.1 |
100.7 |
130.5 |
149 |
X1800 XT |
45.7 |
67 |
97.8 |
123.2 |
135.9 |
X1800 XF |
44.1 |
61.4 |
95.2 |
120.6 |
133.8 |
pixels x fps - Mpixels per second
Crossfire 8x Super AA performance difference  
|
640x480 |
800x600 |
1024x768 |
1280x1024 |
1600x1200 |
X1900 XTX |
-2.50% |
-1.50% |
-3.40% |
-3.40% |
-4.10% |
X1800 XT |
-3.60% |
-8.40% |
-2.70% |
-2.10% |
-1.60% |
percentage
Moving over to OpenGL and our Doom 3 test we see that there is a similar situation as Far Cry with each board rendering 4x FSAA as the performance difference between the single XTX rendering 4x FSAA and two X1900's rendering 8x FSAA is only 4% at high resolution. In this instance the X1900 Crossfire system is rendering this Doom 3 test, with the Ultra settings, in 1600x1200 over an average of 80 FPS with 8x FSAA.
Doom3 - 6 samples per board + 16xAF  
|
307200 |
480000 |
786432 |
1310720 |
1920000 |
X1900 XTX |
44.6 |
63.9 |
88.2 |
108.7 |
121.9 |
X1900 XF |
43.4 |
61.9 |
84.9 |
103.5 |
116.5 |
X1800 XT |
44 |
60.7 |
82.3 |
99.7 |
107.9 |
X1800 XF |
42.5 |
59.4 |
80.3 |
97.9 |
104.3 |
pixels x fps - Mpixels per second
Crossfire 12x Super AA performance difference  
|
640x480 |
800x600 |
1024x768 |
1280x1024 |
1600x1200 |
X1900 XTX |
-2.70% |
-3.10% |
-3.70% |
-4.70% |
-4.40% |
X1800 XT |
-3.40% |
-2.20% |
-2.50% |
-1.80% |
-3.40% |
percentage
The situation is a similar case with each board rendering 6x FSAA. Here the average performance of the X1900 Crossfire is still above 60 FPS at 1600x1200 with 12x FSAA. |