I would have thought GPUs are the way forward, especially when you have AMD, Nvidia and Intel all looking into this. The Tesla cards are already out with 220 cores all running at 1.33ghz, and when you think you can have four of those in a box already the mind boddles at what will be possible. Vray RT already works with GPUs, and are currently working to get the code in Vray to also harness this power. To give an idea of the performance boost on older GPU cards, I've seen a demo of a Vray scene that took 3mins to render, being reduced to 8 seconds.
I know that TG 0.9 has been ported onto Tesla based systems through Cuda. Does anyone know if TG2 has any plans in this direction?
Just think, 880 buckets rendering on your screen instead of 4,8 or with the above chip 12. Mmmm, salivating already.