The source explains that in Slice architecture, the GPU is divided into multiple independent physical units called “slices.” Each slice has its own pipeline, cache, and memory.