3 ways to split work across GPUs — data, pipeline, and tensor parallelism
Click anywhere on the visualization to replay the animation