The Texas Advanced Computing Center (TACC), part of the University of Texas, is having trouble supplying enough time on its three-year-old supercomputer to meet the demand from the university's scientists.

The solution to the problem is to replace the cluster, called Ranger, with a new system whose horsepower will be supplied by 100,000 CPU cores, or between four and five times the current number of cores. Getting the new system – which will be called Stampede – up and running is no easy feat.

Tommy Minyard, associate director for advanced computing systems at TACC, says the data center currently housing Ranger will need to get power, cooling and building upgrades. “This is really a huge leap for us,” Minyard said during a presentation at this week's DatacenterDynamics conference in Dallas.

“We're talking about 10 petaflops peak performance.” TACC is planning to bring Stampede online in January 2013. The system will consist of two clusters: a base cluster and an “accelerated co-processor system”.

The base cluster is being built on Intel processors using Sandy Bridge architecture. The co-processor system will use Intel's many-integrated-core (MIC) processors. As Dell and Intel are busy building the cluster, TACC is busy adding 6.5MW of critical power capacity to the facility's existing 3MW and 12,000 sq ft of raised floor to the 6,000 sq ft it has in the present. The team is also replacing existing chillers with more efficient ones.

The system's primary compute and storage components will live in 182 48U cabinets, cooled by APC in-row cooling systems. Power density of the facility supporting Ranger is 30kW per rack. To support Stampede, the infrastructure is being redesigned to exceed 40kW per rack. And it will need that density in 2015, when the system is upgraded from providing 10 petaflops to more than 15 petaflops. Related Stories IBM research to stop power drain in supercomputers US stimulus funds most powerful supercomputer for climate research