I've seen a couple questions and misconceptions, that I thought people might like cleared up.
First, on the OpenCL port issue, the post at Shacknews was a little mistaken. GPU transcoding in Rage was implemented in CUDA, because OpenCL didn't have the necessary features at the time the work began. (the spec wasn't finished) OpenCL was considered later, but the code relied on a feature that only CUDA had at the time to improve performance by 20+%.
As for supported GPUs, it is everything newer than an 8800 GTX which is supported (8800 GTX lacks one necessary feature). The amount it helps depends on how fast the GPU is and how high your resolution is. The 8800 GT mentioned above can actually help somewhat if the rendering resolution is modest, like 1280x720. At the end of the day, it is a decision of how to allocate GPU power. FWIW, an 8800 GT can reliably accelerate vt_maxPPF 16 (a recommended setting for a Core2Duo) on my Core2Duo at 720p and still reliably get 60 FPS. Putting something new and middle of the road like GTX 560 or even a GTX 550 can typically provide more transcode HP than a Core2 can keep up with. (unless the resolution is really high)
In case people are interested, there was actually a presentation describing some of this work and the associated challenges at GTC in 2010:
http://nvidia.fullviewmedia.com/gtc2010/0921-a1-2152.html
I might answer a couple other questions to help clear things up, but please don't be offended if I don't. Also, I'm not here to get into any favorite IHV arguments. I just wanted to help you understand some (IMHO) interesting tech.
Not speaking for my employer, or any partners, yada, yada.
Thanks Evan, for clearing up some misconception about the CUDA implementation.
And feel free to emphasize any specific technical point you'd like to. Technical trivia about new engine technologies are always appreciated 'round these parts.