ASC News from NNSA
who's whoComings & GoingsUpcoming EventsASC HomeASCeNews POCs

October 2006
Return to this issue’s stories

ASC Roadrunner Era Begins at LANL: Near-term capacity with a future hybrid accelerated option

The first components of the Roadrunner supercomputer arrived in Los Alamos on September 25, just two weeks after the contract with IBM was signed. The primary goals for the Roadrunner system are as follows:

  • Providing a large-capacity computing resource for LANL weapons simulations
  • Implementing optional upgrade to petascale hybrid accelerated architectures capable of supporting future LANL workload.
  • Becoming a lead participant in the industry-wide path toward hybrid accelerated computing devices for HPC.

Roadrunner computer

Josip Loncaric, HPC-5, LANL, inspects the predelivery testing of Roadrunner systems at the IBM Rochester, Minnesota site.

The Roadrunner Project has three phases. During Phase 1, already under way, Los Alamos will acquire, install, and deploy more than 81 teraFLOPS of a base capacity system to provide capacity computing cycles in the near term. Phase 2 is a technology refresh and assessment of the final system. Phase 3 consists of an option to populate the entire classified system with cell blades and achieve a sustained 1 petaFLOP benchmark. The advanced architecture hardware and the base system will be combined to create a hybrid computing architecture that has the potential for significant improvements in the price/performance curve to help meet future ASC computing requirements. Roadrunner is the first supercomputer to use this hybrid processor architecture based on both Opteron X64 processors from Advanced Micro Devices (AMD) and the IBM Cell Broadband EngineTM (Cell BE) processing elements. (Cell BE is getting some media attention because it is the processor Sony created for the PlayStation 3.)

Roadrunner architecture

The Roadrunner hybrid architecture.

The final classified system, planned for 2008, will have 76 teraFLOPS of Opteron and IBM Cell Blades directly connected via InfiniBand to Opteron nodes for Accelerated Architecture, yielding ~1.7 petaFLOPS of cell for sustained petaFLOP performance. It will use 6.1 MW of electricity and require 8,000 square feet of space.

 

DOE Privacy Disclaimer | Sandia Privacy Disclaimer

Contact Us

sandia logo Developed and maintained by Sandia National Laboratories for NA-121.2