The DDL algorithms “train” on visual and audio data, and the more GPUs should mean faster learning. To date, IBM’s record-setting 95 percent scaling efficiency (meaning improved training as more GPUs ...