HD3870 Crossfire Scaling: P35 vs X38

Overall Score

Synthetic Testing: 3DMark
Two variants of Futuremark’s popular 3DMark series were chosen for this comparison. First, the 2003 edition was selected since it typically shows the strongest scaling of systems with Crossfire or SLI multi-GPU setups. In addition, the 2006 version was selected as it should more closely resemble modern gaming loads.

3DMark 2003 testing

As you can see from the chart, the single card score in 3DMark 2003 was about 1.5% better for the X38 board than with the P35, which is within the typical run to run variability of 3D Mark. However, when Crossfire is enabled the difference become much more remarkable – a whopping 25% lead for the X38 system compared to the P35 system. The FPS advantage in the 4 tests which make up 3DMark03 ranged from 19% to 27% for the X38 board.

This result was in line with expectations as 3dMark 2003 is running relatively simple graphics at low resolution and extremely high framerates, so the bandwidth limitation of the P35 board was expected to cause a performance hit, which was in fact observed. Next we’ll look at the performance in 3DMark 2006 which uses more complex graphics at higher resolution.

3DMark 2006 testing

From the chart, we see that the performance hit in 3D Mark 2006 is much less severe than in the 2003 version. This gives us some hope that the performance penalty with the P35 chipset might not be too bad in modern games where the speed of the GPUs themselves is more of an issue than the rate at which the CPU can deliver data to them via the PCIe bus.

Speaking of games, at this point you might be thinking "I don’t care about synthetic benchmarks, and you got over 200 FPS in 3DMark 2003 with a single card, who needs Crossfire in that case? I want to play some games!" If that’s what you’re thinking, up next we’ve got some gaming performance to look at and see if the synthetic benchmarks gave us any indication of gaming performance as they’re supposed to.

30,4,Unreal Tournament 3,”

Unreal Tournament 3
Unreal Tournament 3 is the latest in the UT series from Epic Games, and features fast-paced first-person shooter action in a futuristic environment. To test performance in UT3, the UT3Bench program was used to spectate bot matches. Each test was run 5 times for 1 minute each and the results were averaged to account for different behavior by the bots each time. Minimum, maximum, and average framerates were all recorded using Fraps.

With the X38 board we see decent scaling from the use of Crossfire at all resolutions, but nowhere near a 100% improvement like we would like to see. However the framerates in UT3 are acceptably high even with a single card so the game does not need to have too much effort put into multi-GPU optimizations.

Next on the P35 board we see a slight decrease in framerate by enabling Crossfire at 1024×768 compared to single card performance, but at higher resolutions Crossfire results in an improvement, although not as dramatically as on the X38 board.

Comparing the results you can see that as the resolution increases the P35’s performance deficit is closing, since the GPUs themselves are becoming the limiting factor. Overall, UT3 showed the trend that could be expected.


  • Facebook
  • Twitter
  • Myspace
  • Google Buzz
  • Reddit
  • Stumnleupon
  • Delicious
  • Digg
  • Technorati

Leave A Response