DX12 antiquates SLI and Crossfire. These are both essentially legacy hardware workarounds originating from a fairly hair brained idea in the 90's(3dfx had a version of this the Voodoo series used that separated the work load by alternating scan lines) because the operating system side of things hasn't been able to handle multiple GPU's.
Crossfire and SLI both are very poor methods of working, in some ways they're actually worse than the original SLI 3dfx used, which was itself rife with problems before it self antiquated. To avoid artifact issues, they alternate frames between the cards, which adds significant delay as the frames sit in buffers waiting their turn while the other card spits one out. You can get a smoother visual experience with higher frame rates, yet see no benefit in the responsiveness of the system.
Attempts to use split frame rendering instead, which is the optimal method for performance, more along the lines of what 3dfx was doing back in the day, have gone rather poorly, with major tearing issues being a common result. With AFR, both cards have to load up all the assets, which means your dual GPU effectively has the same amount of memory as a single GPU would. In either case, the two cards aren't acting together, one is used as a slave device, receiving it's orders from the master card, and returning the finished product for a single output. It's as inefficient as it sounds, with the software communicating with one device, which then communicates wit the others.
DX12 makes it a direct line to each piece of hardware, bypassing a need for SLI or Crossfire by sending jobs to the specific cards, and making the more performance worthy methods of combining those assets more viable.
Supporting SLI and Crossfire is something you have to actively implement in software. They probably could do it without issues, but it would be a hassle for gains far inferior to what you'd see by upgrading to DX12 and getting that native support in combination with other efficiency gains.
