Rosetta really doesn’t lend itself to GPU computation. There’s been a few attempts to look into it, but they haven’t really gotten anywhere, mainly due to issues with data transfer between the CPU and GPU. You need to do a significant amount of computation in the GPU in order for it to be worthwhile to do the extensive setup required, and the way Rosetta is structured makes that difficult to achieve.
That said, there has been work putting other, “side” protocols under the GPU that have been successful (things like surface area calculations, and ray-casting based docking, for example), but the main workhorses of Rosetta – things like normal scoring, packing, minimization, fragment insertion, etc., – where most of the protocols spend most of their time haven’t been put into GPU usable form, and it doesn’t look likely they will be in the near future.
.
