Could you please shed some light on the following PMU perf counters on the Grace-Grace? The Grace Performance Tuning guide did not contain the necessary abbreviations
What does OCU stand for in nvidia_scf_pmu_0/ocu_0_cmem_rd_access/?
What does dl, ev and wb stand for in nvidia_scf_pmu_0/cmem_dl/wb/ev_access/?
How are the remote counters different from socket counters? i.e. difference between nvidia_scf_pmu_0/remote_socket_rd_access/ and nvidia_scf_pmu_0/socket_1_rd_access/?
What do prb counters signify? nvidia_scf_pmu_0/socket_3_prb_access/
OCU stands for Ordering Control Unit. It is a PCIe interconnect technology. Note that the ocu events have been removed in the most recent release.
dl stands for dataless, ev stands for eviction, wb stands for writeback, prb stands for probe. Note that the dl, ev, and prb events have been removed in the most recent release.
The remote events count accesses to remote sockets.
The nvidia_scf_pmu_0/remote_socket_rd_access/ counts socket 0 rd accesses to all remote sockets while nvidia_scf_pmu_0/socket_1_rd_access/ counts socket 0 rd accesses to socket 1.