Conclusion

Ashes of the Singularity provides an amazing first look at what to expect from the brave new world of DirectX 12 technology. If implemented correctly, mixed GPU scaling can become a reality, although there are a few things to consider.

First of all, if you plan on upgrading your R9 290X with a R9 490X in a few years and expect massive performance gains based on these results, you might be disappointed. We've seen that the slower card will be the defining factor in how well performance scales. Thinking about the possibilities of DirectX 12, though, this could also depend on the game. Ashes simply splits the viewport into two pieces, which I assume to be of equal size. If a game were able to implement a 60/40 split, for example, the GPU load could be distributed accordingly, which should help with scaling. Another possibility could be to separate the rendering pipeline amongst cards. One card could, for instance, render the scene while the other only does post-processing and physics calculation.

As mentioned before, DirectX 12 also promises some improvements in terms of memory usage. Traditional multi-GPU solutions like SLI and CrossFire have both cards perform the exact same memory allocations for textures and geometry - memory doesn't double. With some clever DX12 coding, I can see this becoming a thing of the past. With split-frame rendering, options are limited. Imagine splitting your monitor in half. You could, for instance, come across objects that are only visible on one side, let's say a stone; you would, as such, have to put the stone's geometry and textures on one card. However, move and it's very likely that the stone would end up on the other side of the monitor an instant later, which drastically reduces any potential for memory savings. It seems as though memory savings only become possible once you truly split the work load up by task, i.e. geometry on one GPU and post-processing on the other, which requires developers to put in some extra effort.

DirectX 12 offers no hand-holding here; if the developer doesn't come up with an ingenious way to optimize mixed-GPU performance, there won't be any scaling at all. This makes me wonder about whether there is any incentive for developers or GPU vendors to properly implement it. I have serious doubts any GPU vendor would allocate resources to help developers with GPU scaling so long as people can turn to competing products to boost performance. I also don't think developers who develop for consoles first these days feel like spending a lot of their own time (= money) on such a feature. Recent releases have shown that many games barely make it through the console-porting process, with the console user interface still in place and tons of bugs and performance issues. Let's hope that I am wrong and that upcoming DirectX 12 titles will finally free us from the GPU vendor lock in.