For all the money and manpower behind them, today's multi-GPU schemes don't offer much flexibility. You need the right motherboard, compatible graphics cards, and driver support for whatever game you're trying to play. An Israeli firm called LucidLogix seeks to toss away much of that complexity with the Hydra Engine, a low-power system on a chip that—in theory—will let users combine up to four GPUs from any single vendor on any platform.
Ryan Shrout over at PC Perspective has put together a little writeup about the Hydra Engine. He says the SoC itself resides on the PCI Express bus while a matching driver operates "between the DirectX architecture and the GPU vendor driver." Using proprietary algorithms, the Hydra Engine breaks up geometry rendering, lighting, post-processing, and other work into different "tasks," and then distributes those tasks between different GPUs based on their capabilities.
This approach sidesteps the latency and bottlenecks of conventional multi-GPU rendering techniques, and it supposedly has the benefit of being pretty much GPU-agnostic. Shrout says users can combine graphics cards from different generations (say, a GeForce 6800 GT and a GeForce 9800 GTX) so long as they're all from the same GPU vendor. Best of all, LucidLogix claims scaling is "nearly linear."
Doesn't all this sound too good to be true? Well, maybe a little. LucidLogix seems quite serious, however, and it announced earlier today that Hydra 100-series SoCs are now available for customer validation. The firm will demonstrate its products in action at the Intel Developer Forum this week, and it expects Hydra-based consumer devices to hit the market in the first half of 2009. Hardware makers will be able to place Hydra Engine chips onto not just add-in PCIe cards, but also motherboards and external graphics "pods" (think AMD's XGP).