Carver: IBM iDataPlex
NERSC's next medium-sized scientific computing system will be an IBM
iDataPlex Linux cluster. The system will be made available to NERSC
users in early 2010.
The IBM system, selected in a competitive procurement, provides excellent
performance, good energy efficiency per flop, and a familiar environment
for mid-range parallel applications. It is intended to replace Bassi
and Jacquard.
The system will be named after the American scientist George Washington Carver.
About George Washington Carver
George Washington Carver was
an American scientist, botanist, educator and inventor whose studies and
teaching revolutionized agriculture in the Southern United States. Much of
Carver's fame is related to the hundreds of plant products he popularized.
More about George Washington Carver on Wikipedia.
Carver Configuration
Complete information on the software configuration, including batch queue
specifications, will be announced in the near future.
| Compute Nodes |
- 400 nodes
- 2 quad-core Intel Nehalem 2.67 GHz processors per node
- 8 cores per node (3,200 total cores)
- 24 GB DDR3 1333 MHz memory per node
- System theoretical peak performance of 34.2 TFlops/sec
|
|
| Login Nodes |
- 8 nodes
- 2 quad-core Intel Nehalem 2.67 GHz processors per node
- 8 cores per node (64 total cores)
- 48 GB DDR3 1066 MHz memory per node
|
|
| Network/Service Nodes |
- 2 nodes
- 2 quad-core Intel Nehalem 2.67 GHz processors per node
- 8 cores per node (16 total cores)
- 48 GB DDR3 1066 MHz memory per node
|
|
| High Performance Interconnect |
- 4X QDR InfiniBand, fibre optic cables
- Local fat-trees with a global 2D mesh
|
| I/O Subsystem |
- NERSC Global Filesystem (NGF) used for global homes, scratch, and project
- NGF scratch will have 1.2 PB of disk space, and provide
peak I/O bandwidth 25 GB/s
|
| User Environment |
- Compilers
- Portland Group Fortran, C, and C++
- GNU Fortran, C, and C++
- UPC
Programming Models
Math Libraries
- MKL (Intel's Math Kernel Library)
- NAG
- PETSc
Development and Performance Tools
- Parallel Debugger TBD
- PAPI Performance API
- IPM (Integrated Performance Monitoring)
NERSC-Provided Applications
- Science domain applications for chemistry, materials sciences
- I/O libraries
- Visualization and analysis tools
- Grid and HPSS software
|
| Usability Features |
-
Fully featured Linux OS on every node.
- Full compute node connectivity with the NERSC Global Filesystem.
- Load-balancing login node selection.
|
| Cooling |
|
| 
