Does anyone know if it's possible to split this model across multiple (identical, sli) GPUs, to run a version bigger than fits in the RAM of a single GPU?