LucidLogix alleged last month that it had developed a solution for combining graphics cards that allowed almost perfect scaling: a pure geometric curve of performance vs the number of discrete GPUs. ExtremeTech previewed this at the ongoing Intel Developers Forum and came away impressed.
There are a couple of problems that both common implementations of combining discrete gpus share: issues with multiple monitors, wildly varying game support and perhaps most worrying of all, poor scaling with additional cards. The company claims that their solution is utterly technology agnostic. It doesn’t matter what graphics platform, driver, company or GPU you use. They all work with the HYDRA engine.
HYDRA can be deployed as in add-in card or as part of the original motherboard. It works by intercepting the instructions sent to the individual cards. The RISC-based System-on-Chip design runs the tasksets through a set of proprietary algorithms. The result is a consummate load balancing solution that allows a much more effective use of individual GPUs. ExtremeTech noted that the small chip allowed two NVIDIA 9800GT video cards achieve frame rates of 45-60fps with all details upped in Crysis.
HYDRA works with both OpenGL and DirectX, and is limited to four cards at this time. While varying models (such as a GTX280 and a 9800) could work together, cards from ATI and NVIDIA won’t work together at this time — this is presumably more a result of OS limitations rather than hardware. LucidLogix is in discussion with board partners and expects to launch with consumer available products coming sometime in 2009.
ExtremeTech article on HYDRA.
LucidLogix page describing the technology.