I know llama.cpp can, it certainly improved performance on my RAM-starved GPU.