AMD has come good on promises that GPU-accelerated Havok physics would be demonstrated at GDC, showing off Havok Cloth (shown in the video below) running on ATI hardware.
Havok Cloth was ported to OpenCL enabling it to run on AMD hardware as per ATI Stream programs. One advantage of using OpenCL is that Havok switch between GPU and CPU on-the-fly, as shown by AMD. Additionally, given that OpenCL, GPU-accelerated physics should be possible without any 'extra' coding over what's needed for Havok on the CPU - as both processors can run the same code - Havok shouldn't find it too hard getting developer support.
On a more disappointing note, AMD wouldn't talk about any performance boost - or lack thereof - as GPU-based Havok is currently only a proof of concept, rather than a completed implementation.
Interestingly, there's nothing to stop an OpenCL version of Havok's physics engine running on nVidia graphics hardware - albeit in competition to PhysX. We'll have to wait and see what happens on that front.