Does this model support multi-GPU workflow? Has anyone tested it? If so, could you please share your valuable experience?