nVidia could be preparing something rather exciting for the 590 launch.

* Unified Virtual Addressing
* GPUDirect v2.0 support for Peer-to-Peer Communication
* Share GPUs across multiple threads
* Use all GPUs in the system concurrently from a single host thread
These are from the release notes for nVidia's BETA Release 270 CUDA 4.0 Candidate driver for developers and select reviewers. It appears that nVidia are making CUDA architecture more parallel, I wonder how this will effect the world of SLi and DirectX11 thread lists?

Also nVidia recently posted this as an announcement on their official forums.

Furmark is an application designed to stress the GPU by maximizing power draw well beyond any real world application or game. In some cases, this could lead to slowdown of the graphics card due to hitting over-temperature or over-current protection mechanisms. These protection mechanisms are designed to ensure the safe operation of the graphics card. Using Furmark or other applications to disable these protection mechanisms can result in permanent damage to the graphics card and void the manufacturer's warranty.
Does make you wonder if a rather power hungry multi GPU is coming rather soon doesn't it?

John