Teraflops in Germany: Microsoft HPC Server Scores High Marks

EddieC 0 309 Views

Less than a year after launching its high performance computing strategy, Microsoft finds its HPC Server 2008 near the top of the heap. The Windows Server 2008 derivative placed 23rd on a list of the 500 top-performing supercomputers in the world, as measured by the National Center for Supercomputing Applications (NCSA). And it’s which is still in beta; general availability is set for the end of this month.

Running on a Dell PowerEdge Cluster equipped with 9,474 processor cores and 9600 GB of memory, HPC Server 2008 performed 68.5 trillion floating point operations (teraflops) per second, according to a Microsoft report. The company was apparently being modest. The NCSA itself in showed , that “Abe” performed at a peak of 88.3 teraflops.

The International Supercomputing Conference publishes a list twice a year at www.top500.org. The news was reported by Microsoft yesterday from the in Dresden, Germany this week. The conference ended today.

All the giants were there: Cray, Fujitsu, HP, IBM, NEC, SGI and Sun, to name just a few. And of course Intel, which in April teamed up with supercomputer maker Cray, and whose Threading Building Blocks form the basis for high performance computing applications the world over.

HPC Server 2008 is based on Windows Server 2008, with the special sauce of failover protection, cluster management, and the ability to scale to thousands of processor cores. HCP Server also includes high-speed networking, supports third-party clustering file systems and has a job scheduler redesigned to perform as part of an SOA.

The fastest system this year was “Roadrunner,” a system built by IBM for the U.S. Department of Energy’s Los Alamos National Laboratory. Its peak performance was a record 1.026 petaflops.

Dani AI

Generated

Thanks to for the original ISC/Windows HPC Server note and to for flagging the conference angle. The important takeaway for database people is this: teraflops measure floating‑point throughput on dense linear algebra tests, not real database performance. That distinction changes what you should evaluate when considering HPC hardware or an HPC stack for SQL Server or other RDBMS workloads.

Practical, workload-focused approach:

  1. Capture and replay a representative workload (or use a DB-focused benchmark such as a TPC-style test or HammerDB) rather than relying on LINPACK-style numbers.
  2. Measure the metrics that matter for DBs: single‑thread latency, memory bandwidth and NUMA behavior, IOPS and latency to storage, and interconnect latency (RDMA/InfiniBand vs Ethernet).
  3. Run the workload on candidate configurations and iterate: tune the OS, storage layout, and DB settings, then re-test.

Architectural checklist to reduce surprises:

  • Prefer low‑latency, RDMA-capable interconnects if you scale across nodes.
  • Confirm the database vendor supports the chosen storage model or clustered filesystem and the HA pattern (shared‑nothing scale‑out vs shared‑storage failover).
  • Plan backup/restore, licensing per core, and operational monitoring up front. Track DB‑specific counters (buffer/cache hit ratios, page life expectancy, I/O queue depth, wait types) during tests.

The ISC result shows Windows can reach high raw FLOPS, but production database decisions should be driven by representative benchmarks, vendor support, and storage/network design rather than teraflops alone.

Member Avatar for Aamy1980
Aamy1980 0 Newbie Poster

That was really a great information about the HPC. Thanks for sharing it here. By the way have you heard about By the way here is a great information about the Conference which is more useful for the Students as well as Job seekers. Have you heard about Cloud Computing Conference 2010 which is the world's largest conference on Cloud computing and its technologies. I got a wonderful opportunity to meet and talk with the world's leading experts of Cloud Computing. The information which I have gathered from the conference helped me to implement in my studies and in my job.

Be a part of the DaniWeb community

We're a friendly, industry-focused community of developers, IT pros, digital marketers, and technology enthusiasts meeting, networking, learning, and sharing knowledge.

Edit Preview