I am not part of Chimera's team, though I have work with it in the past. From my experience I belive it is possible yet difficult to do. If I was tackling this problem I would start with a CUDA tool kit first or openCl. It make take some C knowledge to get the objects to interact nicely. I would also verify the accuracy of you GPU model via running a CPU module of the same object, if when you blast them together you get an identical structure ie. High alignment score you should be able to validate the accuracy of your gpu ran structure and start scale up from there. 

Hope this is helpful or leads you down the right path! I'd be intrested to hear what you find.

Thank you, 

Dylan Ramirez