A good talk at Google I/O that shows how shared memory can reduce shader work in a WebGPU context: https://youtu.be/K2JzIUIHIhc?t=1329