Recount of the (almost) available systems

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 POWER7
    3. IBM BlueGene/Q processor
    4. Intel Xeon
    5. The SPARC processors
  8. Accelerators
    1. GPU accelerators
      1. ATI/AMD
      2. nVIDIA
    2. General computational accelerators
      1. Intel Xeon Phi
    3. FPGA accelerators
      1. Convey
      2. Kuberre
      3. SRC
  9. Interconnects
    1. Infiniband
Available systems
  • The Bull bullx system
  • The Cray XC30
  • The Cray XE6
  • The Cray XK7
  • The Eurotech Aurora
  • The Fujitsu FX10
  • The Hitachi SR16000
  • The IBM BlueGene/Q
  • The IBM eServer p775
  • The NEC SX-9
  • The SGI Altix UV series
  • Systems disappeared from the list
    Systems under development

    In this section we give a recount of the types of systems discussed in the former section and that are marketed presently or will appear within 6 months from now. When vendors market more than one type of machine we will discuss them in distinct subsections. So, for instance, we will discuss IBM systems under entries BlueGene/Q, and eServer p775 because they have a very different structure.

    As already remarked in the Introduction we will not discuss clusters here but restrict ourselves to "integrated" parallel systems. However, the distinction between clusters and integrated systems becomes less and less clear as the network speed of clusters and integrated systems are (almost) comparable and the number of cores in a cluster node also approaches (or in the case of the Fujitsu FX10 is larger than) the number found in the integrated systems. The criteria we use consist of the special measures a vendor has taken to more tightly integrate the nodes in the system than what is found in the standard cluster. Such measures may be in hardware, like special barrier registers, or software, like a vendor-optimised fast intra-node MPI, or both. It may well be that even these kinds of distinctions will disappear over time. In that case the difference will have disappeared completely and we will add the (former) clusters to this overview. Still, we want to mention a few systems here that do not fall within the criteria maintained in the regarding clusters because they represent important systems but cannot be described easily in the usual terms for one or more of various reasons. So, below one will find a few entries that give some information about such configurations but cannot always provide the details one should like to give. Just because we want the reader to be aware of these systems we include them. Because the criteria are so vague, we cannot claim consistency in this respect.

    A fair number of vendors now include GPU-enhanced systems in their portfolio. If sufficient information is available on these systems we also will discuss them, although it will be often impossible to provide reliable performance results for such systems, mainly because of the narrow range of algorithms where they can usefully be applied. If, at all credible we will quote the theoretical peak performance as given by the vendors for these machines.

    A reservation with respect to the word "available" in the heading of this section is in order: rather theoretically available is a better description of the situation. This has nothing to do with the technical ability of the vendors to produce the systems described but everything with the ability of the vendor to invest in selling the system: when a large system is purchased this ought to come with the required maintenance and support. Some vendors cannot or will not sell a system in certain parts of the world because they are not willing or able to provide the necessary maintenance and/or support for the system that is theoretically available. For instance, it is not possible to buy a large Hitachi SR16000 in Europe. Nevertheless we still discuss these systems in the section below for general interest and for those that are located in a part of the world where these vendors deem the selling of their systems economically viable.

    System descriptions

    The systems are presented alphabetically in the pages listed in the sidebar. The "Machine type" entry shortly characterises the type of system as discussed in the former chapter: Processor Array, ccNUMA, etc.