Radeon HD 4870 X2 And HD 4870 X2 CrossFire (4CF) Compared
In the following tables we will show you where the 5% performance increase comes from with two dual-chip cards with HD 4870 X2 (CrossFire 4CF) over one dual-chip card.
The single-chip Radeon HD 4870, two of which can be combined via CrossFire, is labeled with the CF abbreviation. The dual-chip card carries an X2 in its name, which points toward the internal CrossFire link of two graphics processors. When combining an X2 dual-chip card with a single-chip card, you will get 3CF, CrossFire acceleration with three graphics chips. When combining two X2 dual-chip cards via CrossFire, you’ll have four graphics chips running concurrently, which is then labeled as 4CF.
|Radeon HD 4870||1|
|Radeon HD 4870 CF||2|
|Radeon HD 4870 X2||2|
|HD 4870 + HD 4870 X2 3CF||3|
|HD 4870 X2 + HD 4870 X2 4CF||4|
With Assassin’s Creed, the test cards are limited by the CPU—we can only guess how far the values could possibly go up with more processing horsepower. The little hiccup of 4.8% is not an error in the measurement, as we repeated the test several times. When the CPU is limiting frame rates, the results are not surprising. Essentially, the combination that uses the CPU least wins. Four graphics chips don’t give any advantage by overclocking the CPU, since even two graphics chips don’t have enough processing muscle to show off their best numbers.
|Assassins Creed in fps||12800xAA||16800xAA||19200xAA||12804xAA||16804xAA||19204xAA||Overall|
|HD 4870 X2 4CF 8.10 QX@3.67||65.7||70.1||65.6||51.4||50.0||51.9||354.7|
|HD 4870 X2 8.10 QX@3.67||65.4||66.9||67.5||51.3||52.2||52.8||356.1|
|Performance difference in percent||0.5||4.8||-2.8||0.2||-4.2||-1.7||-0.4|
At 1920x1200 with anti-aliasing (AA) the advantage is no less the 40%. The 4CF combo will need higher resolutions to see even larger differences.
|Call of Duty 4 in fps||12800xAA||16800xAA||19200xAA||12804xAA||16804xAA||19204xAA||Overall|
|HD 4870 X2 4CF 8.10 QX@3.67||249.5||250.6||252.7||248.1||243.7||249.4||1494.0|
|HD 4870 X2 8.10 QX@3.67||276.2||265.4||226.5||239.3||199.9||177.1||1384.4|
|Performance difference in percent||-9.7||-5.6||11.6||3.7||21.9||40.8||7.9|
Crysis and the overclocked test CPU are at their limits, so the additional graphics chips of the 4CF combo just draw down extra system power.
|Crysis High in fps||12800xAA||16800xAA||19200xAA||12804xAA||16804xAA||19204xAA||Overall|
|HD 4870 X2 4CF 8.10 QX@3.67||33.2||30.2||30.2||31.0||28.6||28.2||181.4|
|HD 4870 X2 8.10 QX@3.67||34.2||31.1||31.2||31.8||29.1||29.1||186.5|
|Performance difference in percent||-2.9||-2.9||-3.2||-2.5||-1.7||-3.1||-2.7|
Even at very high quality you don’t see a difference. Either the CPU power is not sufficient, or AMD’s drivers are horribly unoptimized in this one. We have heard suggestions from AMD that changes made to 8.11 improve performance, but in our experience with the beta driver, Crysis numbers didn’t budge at all.
|Crysis Very High in fps||12800xAA||16800xAA||19200xAA||12804xAA||16804xAA||19204xAA||Overall|
|HD 4870 X2 4CF 8.10 QX@3.67||28.3||25.4||25.1||26.8||23.8||22.5||151.9|
|HD 4870 X2 8.10 QX@3.67||29.4||26.3||25.7||27.4||24.4||23.8||157.0|
|Performance difference in percent||-3.7||-3.4||-2.3||-2.2||-2.5||-5.5||-3.2|
The result in Quake Wars was expected because similar results of single- and dual-chip cards have shown that one graphics chip is enough to play the game smoothly.
|ET: Quake Wars in fps||12800xAA||16800xAA||19200xAA||12804xAA||16804xAA||19204xAA||Overall|
|HD 4870 X2 4CF 8.10 QX@3.67||86.4||83.8||83.2||83.1||82.6||82.7||501.8|
|HD 4870 X2 8.10 QX@3.67||90.1||87.6||89.1||86.4||86.3||86.1||525.6|
|Performance difference in percent||-4.1||-4.3||-6.6||-3.8||-4.3||-3.9||-4.5|
This is also the case in Half Life 2. One graphics chip is enough for this game, more than that would be a waste with these test settings.
|Half Life 2 Episode 2 in fps||12800xAA||16800xAA||19200xAA||12804xAA||16804xAA||19204xAA||Overall|
|HD 4870 X2 4CF 8.10 QX@3.67||128.4||119.8||120.1||125.9||121.1||120.7||736.0|
|HD 4870 X2 8.10 QX@3.67||130.3||123.5||119.9||128.6||123.4||122.4||748.1|
|Performance difference in percent||-1.5||-3.0||0.2||-2.1||-1.9||-1.4||-1.6|
Using Mass Effect it is easy to test whether several graphics chips are working at all. The upgrade from the single-chip card HD 4870 to the X2 result in a performance gain of 80% and the 4CF combo gave us another 76% on top of that. The developers of the UT3 engine know what they’re doing. Scaling with the number of graphics chips is simply gigantic. Although the CPU power has some influence, four graphics chips still run nicely when other 3D engines have already given up.
|Mass Effect in fps||12800xAA||16800xAA||19200xAA||12804xAA||16804xAA||19204xAA||12808xAA||16808xAA||19208xAA||Overall|
|HD 4870 X2 4CF 8.10 QX@3.67||129.8||126.8||124.3||126.0||123.0||120.5||125.4||123.7||121.5||1121.0|
|HD 4870 X2 8.10 QX@3.67||131.2||129.7||125.0||116.2||87.6||70.6||114.1||85.7||69.0||929.1|
|Performance difference in percent||-1.1||-2.2||-0.6||8.4||40.4||70.7||9.9||44.3||76.1||20.7|
Interesting here, but expected. No win, no loss, the additional graphics chips simply get ignored.
|MS FlightX SP2 in fps||12800xAA||16800xAA||19200xAA||12804xAA||16804xAA||19204xAA||Overall|
|HD 4870 X2 4CF 8.10 QX@3.67||29.5||29.7||29.5||29.8||29.8||29.7||178.0|
|HD 4870 X2 8.10 QX@3.67||29.1||29.7||29.4||29.8||29.8||29.3||177.1|
|Performance difference in percent||1.4||0.0||0.3||0.0||0.0||1.4||0.5|
When looking at the constant results at different resolutions, you will see the CPU limitations for both cards. When two graphics chips don’t get enough data, four won’t do the job any better. Only with a stronger CPU will you get more performance.
|World in Conflict in fps||12800xAA||16800xAA||19200xAA||12804xAA||16804xAA||19204xAA||Overall|
|HD 4870 X2 4CF 8.10 QX@3.67||52.7||53.6||52.6||51.4||51.9||52.3||314.5|
|HD 4870 X2 8.10 QX@3.67||53.5||56.1||54.5||50.0||52.9||52.5||319.5|
|Performance difference in percent||-1.5||-4.5||-3.5||2.8||-1.9||-0.4||-1.6|
When just looking at games with the UT3 engine, 4CF pays off at high resolutions, but on the whole, it is clear that available CPUs are too weak to support four fast graphics chips well enough to justify the extra cost, heat, and power. The gain in performance will increase with the next hardware generation. Until then, the fastest graphics chips will always perform similarly to the pair of chips you get on a single X2 board. Maybe it is even possible to run 16x anti-aliasing with the 4CF combo, but it’s up to you to decide if this really pays off in high resolutions.