Test: 3D performance
[graph 717]
In addition to the issues already mentioned, regarding getting reliable results from the two other games we normally use on the comparative platforms, it almost seems pointless to benchmark more than one game when comparing the performance of motherboards. Invariably the differences in performance, at the high resolutions most people buying this class of motherboard are likely to run their games at, is very small at best.
Even at 1,024x768 in Company Of Heroes, the difference between the two X58-based boards is within our range of acceptable variance, and when comparing cross-platform (X48 vs. X58), the difference is no more than 2.5 per cent. Put simply, once a game is run at a GPU-bound resolution, additional CPU horsepower is relatively pointless.
[graph 697]
[graph 718]
As the Eclipse SLI doesn't come with enough cables as standard to run 3-way SLI; we decided to test 3-way CrossFireX instead using the two bridge cables supplied. As expected, the best scaling occurs at the ultra-high 2,560x1,600 resolution.
When we looked at the eVGA 790i SLI FTW recently, we saw that on a couple of different X48 boards, with a QX9650 CPU and three Radeon HD 3870s, the CrossFireX scaling from 2-way to 3-way was uninspiring to say the least, resulting in no performance gain at 1,920x1,200, and a miserly 10 per cent at 2,560x1,600.
Taking a look at the scaling here, things have improved much, most likely due to the architectural improvements of Radeon HD 4870 over Radeon HD 3870, and in conjunction with better driver optimisations.
At 1,920x1,200 1-2 GPU scaling is a reasonable 60 per cent, but 2-3 GPU scaling still looking quite poor with an increase of just under 12 per cent. At 2,560x1,600, however, scaling improves to 68 per cent and 21 per cent, respectively, and it compares quite well with 3 x 9800 GTXs running on 790i Ultra SLI, which scales, on average, by around 25 per cent at 2,560x1,600 from 2-3 GPUs. Of course, unless you've got really deep pockets neither solution makes sense financially.
