Supercomputing in the State of Montana
Location:
System Name:
Information: |
MERDI, 65 E. Broadway, Butte, MT, 59701
“Big Sky”
info@rmscinc.org |
Performance:
Blade:
Chassis:
Processor (2x/Blade):
Memory:
Storage:
Operating System:
Other:
|
3.8 TFLOPS (theoretical peak); 3.2 TFLOPS Linpack
42 IBM HS21 XM Blades (with InfiniBand 4X connectivity)
i3 IBM BladeCenter H Chassis (with InfiniBand 4X midplane)
Quad-Core Intel Xeon E5440 (2.83GHz 12MB L2 1333MHz 80w; Harpertown)
2 GB memory per core
146 GB per node; not part of GPFS file system (i.e., local or temporary node stores)
RHEL 5; Microsoft HPC Server 2008 (dual bootable and “bare metal” provisioned)
1 IBM POWER p520, RHEL 5 Linux for POWER
IBM p520 used for Blue Gene and Cell BE cross compilations
|
Remote 3D:
Immersive 3D:
Stereographics:
Dual Purpose:
Parallel Interconnect: |
IBM Deep Computing Visualization’s RVN (Remote Visual Networking)
1 node, 32 GB memory, RHEL 5, Desktop Isolation, NVIDIA Quadro FX5600,
2x/node Dual Core Opteron Processor Model 8222 (3.0GHz 1MB L2 1MB L3 Cache)
IBM Deep Computing Visualization’s SVN (Scalable Visual Networking)
1 application node, 32 GB memory RHEL 5
2 rendering nodes, RHEL 5, IB4X, NVIDIA Quadro FX5600, G-Sync II, 4 DVI ports
2x/node Dual Core Opteron Processor Model 8222 (3.0GHz 1MB L2 1MB L3 Cache)
Yes, enabled for Immersive
Yes, when not in use for visualization can be scheduled (co-opt’ed) for compute
Parallel Interconnect: InfiniBand 4x
|
Raw (TBytes):
Usable (TBytes):
Disks:
Usage:
Filesystems: |
72.0
50.4 (approximate)
96 spindles, 4 Gbps SATA (7.2K 750 GB E-DDM), RAID 6, full fail-over redundancy
Scratch/work only; short-term; no backup, restore or hierarchical data services
IBM GPFS (General Parallel File System); NTFS for Windows HPC Server 2008
|
Parallel Interconnect:
External Access:
Maintenance/Service:
|
InfiniBand 4X, full duplex with redundancy and GE failover
25Mbps (360 Networks); Meet-Me VPN for IBM CoD (Computing on Demand)
Private, out-of-band
|
Queuing System:
Parallel Programming:
Compilers:
Libraries:
Debugging:
Management:
Visualization Interface:
|
Torque and Maui; classes TBD
MVAPICH, MPICH, OpenMPI, OpenMP
GNU FORTRAN, GNU C/C++; Absoft HPC SDK, Intel
IBM XLF and XLC++ for POWER
IBM P/ESSL for Linux on POWER, Intel MKL, open source libs
GDB, Allinea and Intel
IBM xCAT, IBM Director; CRI MOAB; Other provisioners being evaluated (TPM, etc.)
NICE EnginFrame Application Gateway
|
(maximums)
Current (@200V):
Max Power Load:
Weight:
Total Thermals: |
199 A
45559 W
1982 Kg (without coolant)
155,100 BTU/hr
|
In computing, FLOPS (or flops or flop/s) is an acronym meaning FLoating point Operations Per Second. The FLOPS is a measure of a computer's performance, especially in fields of scientific and technical calculations that make heavy use of floating point calculations, similar to instructions per second (or MIPS for commercial computing).
future REsources
Phase I—Upgrades
34 Racks Total
130kVA Power Available
30 Tons of Cooling Available
Raised Floor Plenum Space
Overhead Rack for Data Cabling
Floor Loading at 1800-2100lbs per rack
Phase II—Upgrades
54 Racks Total
400kVA Power Available
70 Tons of Cooling Available
Raised Floor Plenum Space
Overhead Rack for Data Cabling
Floor Loading 1800-2100lbs per rack
|
|
