MPI Performance

Benchmark program: Intel/Pallas MPI Benchmarks

More detailed results of

Overview of Linux Configurations:

Linux System: 4-socket Dual-Core AMD Opteron 8220
2.8 GHz, HT1000, 32x2 GByte DDR2-555		 2-socket Dual-Core Intel Xeon 5160 3.0 GHz, Intel 5000V, FSB1333, DDR2-555 2-socket Quad-Core Intel Xeon 5355  2.66GHz, Intel 5000X, DDR2-666 2-socket Dual-Core Intel Xeon 5160 3.0 GHz
Intel 5000V, FSB1333, 4x2 GByte DDR2-555		 2-socket Quad-Core Intel Xeon 5355 2.66GHz, Intel 5000X, DDR2-666 2-socket Dual-Core Intel Xeon 5160 3.0 GHz
Intel 5000V, FSB1333, 4x2 GByte DDR2-555		 2-socket Single-Core  Intel XeonDP
3.4 GHz		 2-socket Single-Core AMD Opteron
2.2 GH 		 4-socket Single-Core AMD Opteron 848
2.2 GH
32 Gbyte DDR1		 2-socket Single-Core Intel Itanium2  1.3 GHz
4 Gbyte DDR		 2-socket Single-Core Intel Itanium2 1.0 GHz,
12 Gbyte DDR
Network: InfiniBand 4x DDR MyriNet 10G MyriNet 10G MyriNet 2000 InfiniBand 4x DDR InfiniBand 4x DDR InfiniBand 4x SDR InfiniBand 4x SDR InfiniBand 4x SDR InfiniBand 4x SDR MyriNet 2000
NIC: Mellanox
PCI-e 8x ConnectX IB MT25418		 MyriCom
MyriNet 10G PCIe-8A-C		 MyriCom
MyriNet 10G, PCIe-8A-C		 MyriCom
MyriNet 2000, PCI-X; M3F-PCIXD		 Silverstorm
PCI-e 8x; memfree
HCA9000-DDR		 Silverstorm
PCI-e 8x; memfree
HCA9000-DDR		 Mellanox
PCI-e 8x
MT25208 II Ex		 Mellanox
PCI-X 133MHz
MTPB 23108-CE128		 InfiniCon
PCI-X 133MHz
InfiniServ HCA 7104		 Mellanox
MTPB 23108-CE128 PCI-X		 MyriCom
MyriNet 2000 PCI-X
MFM-PCIXD-2
Switch: Voltair ISR9024D-M
IB 4x DDR		 MyriNet
10G-SW16LC-8C		 MyriNet
10G-SW16LC-8C		 none Voltair ISR9024D-M
IB 4x DDR		 Voltair ISR9024D-M
IB 4x DDR		 Mellanox
MTS2400		 Mellanox
MTS2400		 Mellanox
MTS2400		 InfiniCon
InfinIO 3032		 back-to-back
Latency
[µs]:		 1,42 2,67 2,61 2,86 3,06 3,26 3,99 4,91 5,43 6,18 9,23
unidir. Bandwidth (1MB)
[MB/s]		 1310 1122 1138 235 1285 1201 904 740 728 707 230
bidir. Bandwidth (1MB)
[MB/s]		 2150 2120 2249 469 2086 1860 1745 806 775 747 432
Barrier (two nodes)
[µs]		 1,61 3,20 3,08 4,11 6,55 4,80 7,00 8,80 6,10 6,96 12,10
MPI version Ohio State University
mvapich 1.0		 MPICH 1.2.7
MX 1.2.0h-rc1		 MPICH 1.2.7
MX 1.2.0h-rc1		 MPICH 1.2.7
MX 1.1.6		 Ohio State University
mvapich 0.9.9		 Ohio State University
mvapich 0.9.9 gen2		 Ohio State University
MPICH 1.2.6
VAPI 0.94		 NCSA
MPICH 1.2.5
VMI 2.0.1		 Scali
MPIConnect 4.3.2		 Ohio State University
MPICH 1.2.5
VAPI 0.92		 MyriCom
MPICH 1.2.5..10
GM 2.0
Date 05.03.2008 31.01.2007 23.01.2007 03.04.2007 04.07.2007 03.05.2007 2005 2005 2005 2005 2003

 

Performance measurements of clusters with Windows Server:

Windows Server  2-socket Dual-Core Intel Xeon 5160 3.0 GHz
Intel 5000V, FSB1333, 4x2 GByte DDR2-555		 2-socket Dual-Core Intel Xeon 5160 3.0 GHz
Intel 5000V, FSB1333, 4x2 GByte DDR2-555		 2-socket Dual-Core Intel Xeon 5160 3.0 GHz
Intel 5000V, FSB1333, 4x2 GByte DDR2-555		 2-socket Dual-Core Intel Xeon 5160 3.0 GHz
Intel 5000V, FSB1333, 4x2 GByte DDR2-555		 2-socket Dual-Core Intel Xeon 5160 3.0 GHz, Intel 5000V, FSB1333, DDR2-555
Network: InfiniBand 4x SDR Gigabit Ethernet InfiniBand 4x DDR Gigabit Ethernet MyriNet 2G
NIC: Silverstorm
PCI-e 8x; memfree
HCA9000-DDR		 Gigabit-Ethernet
on-board		 Silverstorm
PCI-e 8x; memfree
HCA9000-DDR		 Gigabit-Ethernet
on-board		 MyriCom
MyriNet 2000, PCI-X; M3F-PCIXD
Switch: InfiniCon 5000 IB 4x SDR HP ProCurve 1800-24G Voltair ISR9024D-M
IB 4x DDR		 HP ProCurve 1800-24G none
Latency
[µs]:		 5,8 57 13,2 44 13,5
unidir. Bandwidth (1MB)
[MB/s]		 866 64 268 82 179
bidir. Bandwidth (1MB)
[MB/s]		 875 102 486 76 331
Barrier (two nodes)
[µs]		 6,1 160 16,4 226 15,2
SW versions MS-MPI
Windows HPC 2008 SP1		 MS-MPI
Windows HPC 2008 SP1		 MS-MPI
Windows CCS 2003		 MS-MPI
Windows CCS 2003		 MS-MPI
Windows CCS 2003
Date 15.05.2009 15.05.2009 03.07.2007 03.07.2007 22.08.2007
My Staffweb