Columbia (supercomputer)
| Company type | Supercomputer |
|---|---|
| Industry | Computing / Aerospace |
| Founded | 2004 |
| Headquarters | Moffett Field , United States |
| Products | Supercomputing services for NASA research and simulations |
Columbia was a supercomputer developed by Silicon Graphics (SGI) specifically for NASA and installed in 2004 at the NASA Advanced Supercomputing (NAS) center located in Moffett Field, California. Its name commemorated the crew who perished in the Space Shuttle Columbia disaster. This system enhanced NASA's supercomputing capabilities by a factor of ten, enabling advanced support for scientific research, aeronautics, and space exploration initiatives.
The supercomputer facilitated various missions such as detailed simulations of the Space Shuttle and its launch infrastructure, forecasting hurricane paths, modeling global ocean currents, and studying supernova phenomena.
History
When it was introduced in November 2004, Columbia ranked as the second fastest supercomputer on the TOP500 list, achieving a LINPACK performance of 51.87 teraflops. By June 2007, its position shifted to 13th place.
The original configuration consisted of twenty interconnected SGI Altix 3700 systems, each housing 512 processors across multiple racks and operating on SUSE Linux Enterprise. The processors were Intel Itanium 2 Montecito and Montvale models. In 2006, an upgrade was implemented by NASA and SGI that added four Altix 4700 nodes equipped with 256 dual-core processors. This enhancement reduced both the machine’s physical size and its power requirements. These nodes were linked through InfiniBand SDR and DDR cabling capable of data transfer rates up to 10 Gbps.
The selection of the SGI Altix platform followed prior success with Kalpana—a single-node Altix system containing 512 CPUs named after astronaut Kalpana Chawla. Kalpana became integrated as the initial node within Columbia’s twenty-node architecture.
At its maximum capacity, Columbia comprised 10,240 processors supported by 20 terabytes of RAM, alongside 440 terabytes of online storage managed through SGI’s CXFS filesystem. Additionally, it maintained an archival tape storage system with a capacity of 10 petabytes. The collaborative team from NAS, SGI, and Intel received the Government Computer News Agency Award for Innovation in 2005 for deploying Columbia’s processors within a rapid timeframe of 120 days.
Over time, Columbia was systematically retired as newer systems such as the petascale Pleiades and shared-memory Endeavour expanded NASA’s top-tier computational resources. Upon its decommissioning in March 2013, Columbia consisted of four nodes distributed over more than 40 SGI Altix 4700 racks powered by Intel Itanium 2 Montecito and Montvale processors totaling 4,608 cores. Its theoretical peak performance reached 30 teraflops with a memory capacity of 9 terabytes.