Fujitsu Launches PRIMEPOWER HPC2500 Supercomputer

8/26/2002 - Fujitsu Limited released a new massively parallel scalar supercomputer, the PRIMEPOWER HPC2500, which boasts the world's highest theoretical peak performance (up to 85.1 teraflops1) and scalability (maximum of 128 x 128-processor nodes2, or 16,384 processors). Targeted at the High-Performance Computing (HPC) market, the new supercomputer becomes the top-of-line model in Fujitsu's flagship PRIMEPOWER servers for the UNIX market.

PRIMEPOWER HPC2500 uses a massively parallel scalar design based on 1.3 GHz SPARC64TM V processors to achieve the world's highest theoretical peak performance. With this blazing processing performance, for example, the time required for carrying out automotive crash analysis is greatly reduced. Fujitsu's own previous HPC server, the vector-type VPP5000 Series, took roughly 8.2 hours to complete the required calculations, while the new HPC2500 took only 1.1 hours, making it a solid seven times faster3. Another notable potential area of use is in protein folding calculations for sequenced human genome research, which could be reduced eight-fold, from one year to about one and a half months4.

In developing the new supercomputer, Fujitsu capitalized on its established track record and strengths in vector-based supercomputers, including cutting-edge processor technology, parallel execution of calculation instructions and ultrafast interconnect technology, and combined these features with shared memory and memory access control technologies developed for its PRIMEPOWER line of servers. As a result, Fujitsu was able to achieve a theoretical peak performance of 665.6 gigaflops5 per 128-processor node and, when 128 of these nodes (16,384 processors total) are clustered together using its high-speed optical interconnect, an astounding 85.1 teraflops overall. Both of these figures represent record-breaking theoretical peak performance levels. Compared with the VPP5000 Series, the new PRIMEPOWER HPC2500 delivers approximately 70 times higher performance per node, and 17 times higher overall (maximum system configuration of 16,384 processors).

To ensure a smooth transition for customers moving from the VPP Series to the new PRIMEPOWER HPC2500, Fujitsu is offering a parallel compiler (XPFortran) that encapsulates the VPP Fortran specification. The company is also providing industry-standard Fortran and C/C++ compilers that are designed to get the most out of the SPARC64 V processors through cache optimization and other functions, as well as enable high-speed operation simply by recompiling existing programs. Fujitsu also assists customers in migrating to PRIMEPOWER HPC2500 by providing tuning services and other transition-support services.

The new supercomputer runs on the SolarisTM operating environment, for which a large number of applications are commercially available.

Fujitsu's leading edge HPC products are playing a key role in a wide range of scientific and commercial applications, from materials development based on nanotechnology and post-genomic pharmaceuticals, to large-scale scientific calculations in the aeronautics, space, weather and environmental fields, as well as design and simulations for product development in the automotive, heavy industry and electronic equipment industries. Fujitsu will continue to stay at the vanguard of supercomputing technology, and the new PRIMEPOWER HPC2500 promises to make an even greater contribution to supporting leading edge R&D, accelerating and improving advances in science and technology, and improving competitiveness in a variety of industries.

PRIMEPOWER HPC2500 Principal Specifications CPU Processor SPARC64 V
Clock speed; theoretical peak performance 1.3 GHz; 5.2 GFLOPS
Node Number of processors 8-128 CPUs per node
Main memory max 512 GB per node
System High-speed optical interconnect 2 channels of max.16 GB/sec (in, out) per node
Maximum system configuration
(CPUs; theoretical peak performance) 128 nodes
(16,384 CPUs; 85.1 TFLOPS)

Delivery: From January 2003
Sales target: 100 systems over next two years
Pricing: 167.18 million yen (Japan) for PRIMEPOWER HPC2500 with 8 CPUs, 2GB memory, and 18GB hard disk)

About Fujitsu (
Fujitsu is a leading provider of customer-focused IT and communications solutions for the global marketplace. Pace-setting technologies, high-reliability/performance computing and telecommunications platforms, and a worldwide corps of systems and services experts make Fujitsu uniquely positioned to unleash the infinite possibilities of the broadband Internet to help its customers succeed. Headquartered in Tokyo, Fujitsu Limited (TSE:6702) reported consolidated revenues of 5 trillion yen (about US$38 billion) for the fiscal year ended March 31, 2002.

All SPARC trademarks are registered trademarks of Fujitsu in the USA and other countries, under license from SPARC International Inc of the USA. UNIX is a registered trademark of The Open Group in the USA and in other countries. SPARC64 is a trademark of Fujitsu used under license from SPARC International, Inc. of the USA. Sun, Sun Microsystems, the Sun logo, Solaris, and all Solaris-related trademarks and logos are registered trademarks of Sun Microsystems in the USA and other countries, and are used by Fujitsu under license.

*1. Teraflops (TFLOPS):
One trillion floating point operations per second-a measurement of computing performance.
*2. Node:
a computing unit that shares a single memory space.
*3. Automotive crash analysis:
Comparison of maximum system configurations for both the VPP5000 Series and the PRIMEPOWER HPC2500. Comparison of simultaneous processing of 128 jobs (128 study cases) based on conceptual model involving 280,000 elements and a physical time of 160msec.
*4. Protein folding calculations for sequenced human genome research:
Comparison of maximum system configurations for both the VPP5000 Series and the PRIMEPOWER HPC2500. Computational model using molecular dynamics simulation, 10,000-20,000 atoms, 1010 calculation steps.
*5. Gigaflops (GFLOPS):
One billion floating point operations per second-a measurement of computing performance.

Previous Page | News by Category | News Search

If you found this page useful, bookmark and share it on: