How Does It Work?
Generally speaking, SLI and CrossfireX work by dividing a frame up into portions and rendering each portion on a separate card or by alternating the rendering load between cards. These are reasonably efficient methods, but its easy to see why they don't allow for compatibility between different models or brands. If one were to pair a Radeon 5870 with a Radeon 4870, split a frame up, and then give one half to each card, the Radeon 5870 would be done first. The Radeon 5870 would therefor be limited by what the Radeon 4870 could do.
Lucid's Hydra chip works differently. Instead of simply splitting frames into portions, it intercepts API calls, takes a look at them, and organizes them in the way it best sees fit. Then it distributes that information evenly across the GPUs in the system. In other words, the Hydra chip is actually intercepting the data that normally goes to the GPU and its driver and re-distributing that data for maximum efficiency. Lucid has proven this technology works by displaying the work done by two different GPUs on two different monitors. The image rendered was changing dynamically from frame to frame based on how the Hydra felt the information should be most efficiently rendered. Of course, the chip would normally recombine that information into one image which would then be displayed.
What is incredible about this technology, besides the fact that it allows GPUs of any model or brand to work together, is that it should allow for any number of GPUs to work together. This includes not only quad-GPU setups, which are the current maximum supported by motherboards, but also tens or even hundreds of GPUs working in tandem. Intel's interest in Lucid may be tied to Intel's graphics card project, Larrabee, which uses large numbers of processors on one card.