there is a great summary of CGA or "cluster_group" in CG. but i'm curious about cluster (for TPC or GPC?) there is no details found in offical docs. for hardware implementation, the distribute shared memory seems more easily shared inner TPC for physical or hardware limitation. if cluster is GPC. the communication (distribute) between TPC (the basic partition in physical design) seems not easy? or the distribute shared memory cross TPC is fake? (use global memory sync?)