IBM makes big showing at SC12 with extensive HPC innovations

Award-winning presence highlights IBM’s participation at supercomputing conference

Email the article  Download pdf

TC-Insights-SC12IBM’s presence was certainly felt at the recent information-packed SC12, the International Conference for High Performance Computing, Networking, Storage and Analysis. At the event in Salt Lake City, Utah from November 10–16, 2012, IBM won 6 of the top 10 positions in the TOP500 list of the world’s fastest supercomputers. IBM® Platform Computing™ won the Intel Pathfinder Award for leading the development of Intel Cluster Ready Reference Designs for the E5 family of products. HPCwire awarded IBM a number of prestigious Reader’s Choice and Editor’s Choice Awards.

Other developments included the demonstration of game-changing cloud solutions for high-performance computing (HPC) and announcements about innovations that support the recently released Intel® Xeon® Phi™ coprocessor system. IBM also showcased the latest in storage and memory solutions, as well as advances in tools such as IBM Parallel Environment Developer Edition 1.2 and the open-source Eclipse Parallel Tool Platform.

IBM servers took 6 of the top 10 positions in the new TOP500 list of leading supercomputing sites announced at SC12. All three IBM HPC technologies—Blue Gene/Q®, x86 System x® and POWER7®—are represented in this impressive ranking: 

  • Number 2: The Sequoia supercomputer at Lawrence Livermore National Laboratory (Blue Gene/Q systems)
  • Number 4: The Mira installation at the U.S. Department of Energy’s Argonne National Laboratory (Blue Gene/Q systems)
  • Number 5: The Juqueen supercomputer at the German research institute Forschungszentrum Jülich (Blue Gene/Q systems)
  • Number 6: The SuperMUC supercomputer at the Leibniz-Rechenzentrum (LRZ) computer center in Germany (System x iDataPlex®)
  • Number 9: The Fermi installation at the CINECA consortium in Italy (Blue Gene/Q systems)
  • Number 10: The Defense Advanced Research Projects Agency (DARPA) Trial Subset (Power 775)

Overall, IBM systems power 192 sites on the new TOP500 list—more than any other server vendor.

IBM Platform Computing showcased HPC cloud solutions including significant advances in on-demand access to shared pools of configurable resources. Accessing these capabilities through the cloud appears to be a natural evolution in the use of HPC and big data analytics, particularly for engineering and scientific applications that already leverage cluster and grid computing. 

Demonstrations of the Platform Dynamic Cluster, for example, illustrated new ways to improve service levels while allowing organizations to do more work with less infrastructure. By transforming static Platform LSF® clusters into a dynamic cloud infrastructure, the IBM team showed attendees that cluster composition can be automatically transformed to meet changing workload demands.

The IBM team also walked attendees through the time-saving capabilities of the newly enhanced Platform Cluster Manager – Advanced Edition. Through a secure, multi-tenant architecture, Platform Cluster Manager – Advanced Edition enables the consolidation of disparate technical computing and analytics cluster infrastructures, while allowing multiple user groups to still function independently. The software provides the foundation for building a dynamic shared analytics and HPC cloud by enabling self-service creation and management of multiple heterogeneous clusters from a shared resource pool, and the ability to scale a running cluster up or down as the job volume changes.

SC12 also saw Intel announce its Intel Xeon Phi coprocessor designed to deliver more overall performance for HPC applications in which many calculations are carried out simultaneously. To help organizations take advantage of these advances—and achieve faster results from improved system management and infrastructure utilization—the IBM System x iDataPlex dx360 M4 now provides the Intel Xeon Phi coprocessor as an option.

To illustrate these capabilities, representatives from NASA Goddard explained how they used the IBM Intelligent Cluster™ process to ensure that a recent deployment of iDataPlex systems arrived racked, cabled and ready to go for initial testing during Hurricane Sandy. Even with power outages, the Phi coprocessors were easily installed and enabled, and allowed the NASA Goddard team to perform a LINPACK run with Phi coprocessors that now ranks at number 52 on the TOP500 list.

Attendees learned that IBM’s support for Xeon Phi goes even further. IBM Platform HPC and Platform LSF management software is enabled to provide built-in management support for Intel Xeon Phi and automatically allocates the Intel Xeon Phi coprocessors to application workloads suited for those resources. These Platform products also enable administrators to monitor key metrics and optimize utilization of the new Intel coprocessor.


The importance of storage was highlighted at SC12. IBM introduced the new General Parallel File System (GPFS™) Storage Server, which has been designed for immediate integration with big data applications. RAID and disk management functions execute on IBM servers directly attached to hundreds of disks, rather than using traditional external hardware controllers, helpings to improve total cost of ownership with high-performance, nonintrusive disk diagnostics and improved storage affordability. This design delivers extreme end-to-end data integrity and enables faster rebuild and recovery times by reducing rebuild overhead by 3.5 times over current capabilities. Additional features include de-clustered RAID (8+2p, 8+3p), 2- and 3-fault-tolerant parity codes, end-to-end checksum from disk platter to client, and protection against lost writes, off-the-shelf, high-density JBODs, and built-in solid-state drive (SSD) acceleration.

Very large memory processing was showcased, too, by demonstrating the System x3750 X5 servers running ScaleMP. A live TS3500 tape solution paired with the DCS3700 storage let attendees see how end-to-end storage and tape archiving can be a very cost-effective solution for managing the volumes of data generated by HPC processes.

HPCwire awards are widely recognized as the most prestigious recognition given by the HPC community to its own each year. 

Editor's choice awards included:
  • Top supercomputing achievement: The deployment of IBM Blue Gene/Q Sequoia supercomputer at Lawrence Livermore National Laboratory
  • Best use of HPC in life sciences: IBM Watson™ expert system for WellPoint Inc.
  • Best application of “green computing” in HPC: IBM Blue Gene/Q
Reader's Choice award included:
  • Top 5 Vendors to Watch: IBM

If you weren’t able to attend the show, catch up by watching videos of IBM at SC12 or IBM Platform Computing Solutions at SC12.