Blue Gene User Guide
The Blue Gene system is SDSC's largest IBM system, housed in three racks with 3,072 compute nodes (6,144 processors) and 384 I/O nodes. This maximum ratio of I/O to compute nodes was chosen to support data-intensive computing. Each node consists of two PowerPC processors that run at 700 MHz and share 512 MB of memory, giving an aggregate peak speed of 17.2 teraflops (that's 6,144 cores x 2.8 Gflop/s per core = 17.2 Tflop/s) and a total memory of 1.5 TB. Blue Gene has achieved 13.83 teraflops on the widely quoted Linpack benchmark.
To ensure effective parallel processing, all compute nodes are connected by two high-speed networks: a 3-D torus for point-to-point message passing and a global tree for collective message passing. All I/O nodes are connected internally to the global tree and externally via gigabit Ethernet. The peak I/O rate for SDSC's data-optimized configuration is 8 GB/s at present, and 3.2 GB/s has already been demonstrated.
With its large number of processors in a compact footprint, Blue Gene enables reductions in power consumption, cooling, and space requirements for institutions requiring immense computing power. The new architecture's ability to produce cost-effective compute power in such a small package provides a glimpse into the future of supercomputing. Funding has been provided by the National Science Foundation and the Department of Energy.
Recommended Usage Guidelines
To ensure that other users can access the machine, you are requested to follow the guidelines outlined in the Partition Layout and Usage Guidelines section of the Running Jobs page. Blue Gene is primarily intended to run codes that scale very well, have only fine-grained parallelism, and do not use much memory per node.
For code that requires more computing power than is available on BlueGene, please see "Migrating Code to Other NSF Resources."
