SPEC® MPIM2007 Result

Dell, QLogic, ClusterVision,

U. of Cambridge HPC Cluster Darwin,
QLogic InfiniBand Interconnect

SPECmpiM_peak2007 = Not Run

SPECmpiM_base2007 = 6.44

MPI2007 license:	0018	Test date:	May-2007
Test sponsor:	QLogic Corporation	Hardware Availability:	Jul-2006
Tested by:	QLogic Performance Engineering	Software Availability:	Feb-2007

Results Table

Benchmark	Base							Peak
Benchmark	Ranks	Seconds	Ratio	Seconds	Ratio	Seconds	Ratio	Ranks	Seconds	Ratio	Seconds	Ratio	Seconds	Ratio
Results appear in the order in which they were run. Bold underlined text indicates a median measurement.
104.milc	64	286	5.47	281	5.56	283	5.54
107.leslie3d	64	1087	4.80	1009	5.18	1094	4.77
113.GemsFDTD	64	653	9.66	651	9.69	683	9.24
115.fds4	64	352	5.54	352	5.54	345	5.65
121.pop2	64	560	7.37	584	7.07	555	7.44
122.tachyon	64	792	3.53	776	3.60	778	3.59
126.lammps	64	482	6.05	484	6.02	484	6.03
127.wrf2	64	764	10.2	768	10.2	756	10.3
128.GAPgeofem	64	266	7.75	266	7.77	266	7.76
129.tera_tf	64	558	4.96	590	4.69	587	4.71
130.socorro	64	489	7.80	492	7.76	432	8.84
132.zeusmp2	64	488	6.35	487	6.37	488	6.36
137.lu	64	465	7.90	511	7.19	471	7.81

Hardware Summary
Type of System:	Homogeneous
Compute Node:	Dell PowerEdge 1950
Interconnects:	QLogic InfiniBand HCAs and switches Ethernet Network for File Server Access
File Server Node:	Dell PowerVault MD1000
Head Node:	Dell PowerEdge 1950
Total Compute Nodes:	16
Total Chips:	32
Total Cores:	64
Total Threads:	64
Total Memory:	128 GB
Base Ranks Run:	64
Minimum Peak Ranks:	--
Maximum Peak Ranks:	--

Software Summary
C Compiler:	QLogic PathScale C Compiler 3.0
C++ Compiler:	QLogic PathScale C++ Compiler 3.0
Fortran Compiler:	QLogic PathScale Fortran Compiler 3.0
Base Pointers:	64-bit
Peak Pointers:	64-bit
MPI Library:	QLogic InfiniPath MPI 2.0
Other MPI Info:	None
Pre-processors:	No
Other Software:	None

Node Description: Dell PowerEdge 1950

Hardware
Number of nodes:	16
Uses of the node:	compute, head
Vendor:	Dell
Model:	Dell PowerEdge 1950
CPU Name:	Intel Xeon 5160
CPU(s) orderable:	1-2 chips
Chips enabled:	2
Cores enabled:	4
Cores per chip:	2
Threads per core:	1
CPU Characteristics:	1333 MHz system bus
CPU MHz:	3000
Primary Cache:	32 KB I + 32 KB D on chip per core
Secondary Cache:	4 MB I+D on chip per chip
L3 Cache:	None
Other Cache:	None
Memory:	8 GB (8 x 1 GB PC2-5300F)
Disk Subsystem:	SAS, 73 GB, 15000 RPM
Other Hardware:	None
Adapter:	QLogic InfiniPath QLE7140
Number of Adapters:	1
Slot Type:	PCIe x8
Data Rate:	InfiniBand 4x SDR
Ports Used:	1
Interconnect Type:	InfiniBand

Software
Adapter:	QLogic InfiniPath QLE7140
Adapter Driver:	InfiniPath 2.0
Adapter Firmware:	None
Operating System:	ClusterVisionOS 2.1 Based on Scientific Linux SL release 4.3 (Beryllium)
Local File System:	Linux/ext3
Shared File System:	NFS
System State:	Multi-User
Other Software:	Torque 2.1.2

Node Description: Dell PowerVault MD1000

Hardware
Number of nodes:	1
Uses of the node:	file server
Vendor:	Dell
Model:	Dell PowerEdge 1950
CPU Name:	Intel Xeon 5160
CPU(s) orderable:	1-2 chip
Chips enabled:	2
Cores enabled:	4
Cores per chip:	2
Threads per core:	1
CPU Characteristics:	1333 MHz system bus
CPU MHz:	3000
Primary Cache:	32 KB I + 32 KB D on chip per core
Secondary Cache:	4 MB I+D on chip per chip
L3 Cache:	None
Other Cache:	None
Memory:	4 GB (4 x 1 GB PC2-5300F)
Disk Subsystem:	13.5 TB: 3 x 15 x 300 GB, SAS, 10000 RPM 3 Dell PowerVault MD1000 Disk Arrays, each one has 15 disks.
Other Hardware:	None
Adapter:	Chelsio T310 10GBASE-SR RNIC (rev 3)
Number of Adapters:	1
Slot Type:	PCIe x8 MSI-X
Data Rate:	10 Gbps Ethernet
Ports Used:	1
Interconnect Type:	Ethernet

Software
Adapter:	Chelsio T310 10GBASE-SR RNIC (rev 3)
Adapter Driver:	cxgb3 1.0.078
Adapter Firmware:	T 3.3.0
Operating System:	ClusterVisionOS 2.1 Based on Scientific Linux SL release 4.3 (Beryllium)
Local File System:	Linux/ext3
Shared File System:	NFS
System State:	Multi-User
Other Software:	None

General Notes

A separate node handling login and resouces management
is not listed as it is not performance related.

Interconnect Description: QLogic InfiniBand HCAs and switches

Hardware
Vendor:	QLogic
Model:	InfiniPath adapters and Silverstorm switches
Switch Model:	QLogic SilverStorm 9080 Fabric Director (InfiniBand switch)
Number of Switches:	1
Number of Ports:	96
Data Rate:	InfiniBand 4x SDR and InfiniBand 4x DDR
Firmware:	3.4.0.1.3
Topology:	Full Bisectional Bandwidth, Fat-Tree, Max 3 swith-chip hops.
Primary Use:	MPI traffic

General Notes

The 16 nodes used are from one CU (Computational Unit, 65 nodes)
of the 9 CUs in the Darwin cluster. Jobs within one CU
use one SilverStorm 9080 switch.
The data rate between InifniPath HCAs and SilverStorm switches
is SDR. However, DDR is used for inter-switch links.

Interconnect Description: Ethernet Network for File Server Access

Hardware
Vendor:	Chelsio, Nortel
Model:	Chelsio T310 adapters and Nortel 5530 5510 8610 switches
Switch Model:	Nortel Ethernet Routing Switch 5510-24T
Number of Switches:	1
Number of Ports:	24
Data Rate:	1 Gbps Ethernet
Firmware:	1.0.0.16
Switch Model:	Nortel Ethernet Routing Switch 5510-48T
Number of Switches:	3
Number of Ports:	48
Data Rate:	1 Gbps Ethernet
Firmware:	1.0.0.16
Switch Model:	Nortel Ethernet Routing Switch 5530-24TFD
Number of Switches:	2
Number of Ports:	26
Data Rate:	1 Gbps Ethernet (24 ports) and 10 Gbps Ethernet (2 ports)
Firmware:	4.2.0.12
Switch Model:	Nortel Passport 8610 switch 4.1.0.0
Number of Switches:	1
Number of Ports:	24
Data Rate:	10 Gbps Ethernet
Firmware:	Optivity Switch Manager version 4.1
Topology:	Three CUs are connected with six Ethernet Routing switches 5530-24TFD, 5510-24T and 5510-48T as a ring. Each of two 5530-24TFD switches is connected to the Nortel Passport 8610 switch through two 10Gbit ports. See Slide 10 of NortelEthernetSwitchDiagram.pdf for a network diagram.
Primary Use:	file system traffic

Base Compiler Invocation

C benchmarks:

/usr/bin/mpicc -cc=pathcc

C++ benchmarks:

126.lammps:

/usr/bin/mpicxx -CC=pathCC

Fortran benchmarks:

107.leslie3d:	/usr/bin/mpif90 -f90=pathf90
113.GemsFDTD:	/usr/bin/mpif90 -f90=pathf90
115.fds4:	/usr/bin/mpif90 -f90=pathf90
129.tera_tf:	/usr/bin/mpif90 -f90=pathf90
132.zeusmp2:	/usr/bin/mpif90 -f90=pathf90
137.lu:	/usr/bin/mpif90 -f90=pathf90

Benchmarks using both Fortran and C (except as noted below):

/usr/bin/mpicc -cc=pathcc /usr/bin/mpif90 -f90=pathf90

Base Portability Flags

104.milc:	-DSPEC_MPI_LP64
121.pop2:	-DSPEC_MPI_DOUBLE_UNDERSCORE -DSPEC_MPI_LP64
122.tachyon:	-DSPEC_MPI_LP64
127.wrf2:	-DF2CSTYLE -DSPEC_MPI_DOUBLE_UNDERSCORE -DSPEC_MPI_LINUX -DSPEC_MPI_LP64
128.GAPgeofem:	-DSPEC_MPI_LP64
130.socorro:	-fno-second-underscore -DSPEC_MPI_LP64

Base Optimization Flags

C benchmarks:

-march=core -Ofast -OPT:malloc_alg=1

C++ benchmarks:

126.lammps:

-march=core -O3 -OPT:Ofast -CG:local_fwd_sched=on

Fortran benchmarks:

107.leslie3d:	-march=core -O3 -OPT:Ofast -OPT:malloc_alg=1 -LANG:copyinout=off
113.GemsFDTD:	-march=core -O3 -OPT:Ofast -OPT:malloc_alg=1 -LANG:copyinout=off
115.fds4:	-march=core -O3 -OPT:Ofast -OPT:malloc_alg=1 -LANG:copyinout=off
129.tera_tf:	-march=core -O3 -OPT:Ofast -OPT:malloc_alg=1 -LANG:copyinout=off
132.zeusmp2:	-march=core -O3 -OPT:Ofast -OPT:malloc_alg=1 -LANG:copyinout=off
137.lu:	-march=core -O3 -OPT:Ofast -OPT:malloc_alg=1 -LANG:copyinout=off

Benchmarks using both Fortran and C:

121.pop2:	-march=core -Ofast -OPT:malloc_alg=1 -O3 -OPT:Ofast -LANG:copyinout=off
127.wrf2:	Same as 121.pop2
128.GAPgeofem:	Same as 121.pop2
130.socorro:	Same as 121.pop2

Base Other Flags

C benchmarks:

-IPA:max_jobs=4

C++ benchmarks:

126.lammps:

-IPA:max_jobs=4

Fortran benchmarks:

107.leslie3d:	-IPA:max_jobs=4
113.GemsFDTD:	-IPA:max_jobs=4
115.fds4:	-IPA:max_jobs=4
129.tera_tf:	-IPA:max_jobs=4
132.zeusmp2:	-IPA:max_jobs=4
137.lu:	-IPA:max_jobs=4

Benchmarks using both Fortran and C (except as noted below):

-IPA:max_jobs=4

The flags file that was used to format this result can be browsed at
http://www.spec.org/mpi2007/flags/MPI2007_flags.20070717.html.

You can also download the XML flags source by saving the following link:
http://www.spec.org/mpi2007/flags/MPI2007_flags.20070717.xml.