Hum this could be nice for sure, but smode is actually using opengl which is not really multi-gpu oriented api.
Also note that SLI means you duplicate the textures / meshes / etc in 2 video card vram which is not the top usage of multi gpu, it can be even worst because you saturate the PCI express with more message than if you were using a single GPU. (For instance there is probably no benefit to use SLI with smode if you are using a single output because it need to download/upload computed part result in one of the gpu into the other one, which probably take more time than just computing everything on a single gpu).
The best would be that smode allocate ressources by output using the gpu memory of this output, but this is probably requiring switch to another api like vulkan, less compatible with old hardwares, and a lot of developpement time.
OpenGL 3.3 is getting old (You can't even select the gpu you are using with the api, you need to use tricks that are not always working), but it's still the most available API on large range computer / graphics cards.
I'm afraid this is not our top priority for now.
