Intra-node performance numbers of MVAPICH2 on Magny Cours Architecture (05/06/13)

• Experimental Testbed: Each node of our testbed has 24 AMD Opteron 6174 processors running at 2.2 GHz with 512 KB L2 cache. Each node also has 32 Gigabyte memory, x8 PCI Express Gen2 interfaces and Mellanox ConnectX-2 QDR HCAs with PCI Express interfaces in multi-rail configuration. The nodes are connected using a 36 port Mellanox QDR InfiniBand switch with QSFP ports. The operating system used was Red Hat Enterprise Linux Server release 5.5 (Tikanga).

• MVAPICH2 currently delivers put latency of 0.90 microseconds , get latency of 0.90 microseconds and accumulate latency of 1.10 microseconds within the socket for 4 bytes, using active synchronization. Between sockets,it delivers put latency of 1.18 microseconds , get latency of 1.00 microseconds and accumulate latency of 1.25 microseconds for 4 bytes.

• Processes were mapped onto cores 1 and 2 to take the intra socket numbers and onto 1 and 12 to take the inter socket numbers.