The Fujitsu/Siemens PRIMEQUEST 500

Introduction
HPC Architecture
  1. Shared-memory SIMD machines
  2. Distributed-memory SIMD machines
  3. Shared-memory MIMD machines
  4. Distributed-memory MIMD machines
  5. ccNUMA machines
  6. Clusters
  7. Processors
    1. AMD Opteron
    2. IBM POWER5+
    3. IBM BlueGene processors
    4. Intel Itanium 2
    5. Intel Xeon
    6. The MIPS processor
    7. The SPARC processors
  8. Networks
    1. Infiniband
    2. InfiniPath
    3. Myrinet
    4. QsNet
    5. SCI
Available systems
  1. The Bull NovaScale
  2. The C-DAC PARAM Padma
  3. The Cray X1E
  4. The Cray XT3
  5. The Cray XT4
  6. The Cray XMT
  7. The Fujitsu/Siemens M9000
  8. The Fujitsu/Siemens PRIMEQUEST
  9. The Hitachi BladeSymphony
  10. The Hitachi SR11000
  11. The HP Integrity Superdome
  12. The IBM eServer p575
  13. The IBM BlueGene/L&P
  14. The Liquid Computing LiquidIQ
  15. The NEC Express5800/1000
  16. The NEC SX-8
  17. The SGI Altix 4000
  18. The SiCortex SC series
  19. The Sun M9000
Systems disappeared from the list
Systems under development
Glossary
Acknowledgments
References

Machine type RISC-based shared-memory multi-processor
Models PRIMEPOWER 540, 580
Operating system Linux (RedHat EL4 or SuSE SLES 9/10)
Connection structure Crossbar
Compilers Parallel Fortran 90, OpenMP, C, C++ (Intel)
Vendors information web page: http://www.computers.us.fujitsu.com/www/products_primequest.shtml?products/servers/primequest/index
Year of introduction 2006

System parameters:

Model PRIMEQUEST 540 PRIMEQUEST 580
Clock cycle 1.6 GHz 1.6 GHz
Theor. peak performance    
Per proc. core (64-bits) 6.4 Gflop/s 6.4 Gflop/s
Maximal 204.8 Gflop/s 409.6 Gflop/s
Main memory    
Memory/maximal ≤ 1 TB ≤ 2 TB
No. of processors 4—16 4—32
Communication bandwidth    
Point-to-point --- ---
Aggregate 68.2 GB/s 136.4 GB/s

Remarks:

We only discuss here the PRIMEQUEST 540 and 580 as the smaller model, the 520 has the same structure but less processors (maximally 4). The PRIMEQUEST is one of the many Itanium-based machines that are offered these days. The older 400 series that is on the market for less than a year is now replaced by the 500 series that contains the dual-core Montecito variant, or Itanium 9000 as it is named officially. As the clock frequency is 1.6 GHz the peak performance is just over 200 Gflop/s for a model 540 and about 410 Gflop/s for a model 580.

The proprietary network to build clusters from the 540 or 580 nodes are not offered as was the case for the late PRIMEPOWER systems (see Systems disappeared from the list) probably because the PRIMEQUESTs use a standard Linux distribution for an operating system instead of the Solaris variant that Fujitsu employs in the SPARC64-based M9000 systems. So, when one (or Fujitsu) is prepared to build a large configuration based on the PRIMEQUEST one would have to used a third party network like Infiniband, Myrinet, or Quadrics.

The PRIMEQUEST is not a ccNUMA machine but a SMP system where all processors have equal access to the common and potentially large memory. Such a choice dictates a modest amount of processors (32 maximum here) and a high-speed crossbar. The latter requirement is fulfilled by a full crossbar with an aggregate bandwidth of 136.4 GB/s for the PRIMEQUEST 580.

Obviously, the compiler suite provided by Intel is used on the processors and therefore will there be virtually no difference between a single processor of the PRIMEQUEST and other Montecito-based systems. Only in parallel OpenMP and MPI programs the differences should show.

Fujitsu tries to stand out with respect to other vendors in offering fail-safe systems. For instance it is possible to have the crossbar doubled. Would one crossbar fail, the other would take over without interrupt of service thus securing the safe completion of the active tasks. Apart from the second crossbar also other components can be doubled, of course at a price, to make the system highly reliable.

Note Large HPC configurations of the PRIMEQUEST are not sold in Europe as they are judged to be of insufficient economical interest by Fujitsu-Siemens.

Measured Performances:
An ensemble of 10 PRIMEQUEST systems has, according to [49] attained a speed of 3119 Gflop/s for a linear system of size N = 499,200. This amounts to an efficiency of 76%.