The QCDOC, Quantum Chromodynamics on a Chip, is a realized in different places supercomputer concept, which aims to get by cheap and simple but effective hardware a massively parallel supercomputer, which is small in the Ausmessungen and preferably accounts for the quantum chromodynamics ( QCD ) is carried out. So he is also a "cheap special-purpose computer " and, due to the massive parallelism, a highly efficient supercomputer, preferably for QCD. At the same time was very on the energy efficiency of the computer.
The concept was first developed as a joint project of several institutions: the University of Edinburgh ( UKQCD ), Columbia University (New York), the RHIC accelerator facility at Brookhaven National Laboratory (NY ), and the Group, IBM. The aim was to enable highly effective in lattice QCD computer simulations. Should be achieved at least 10 Tflops at a capacity utilization of 50%.
There are three implemented QCDOCs, each aimed at maximum power, (10 Tflops ).
- University of Edinburgh ( Parallel Computing Centre ( EPCC ), in operation since 2005)
- 1 Brookhaven ( RHIC )
- Brookhaven 2 (U.S. Department of Energy) (DOE - high-energy program at the accelerator at Brookhaven. )
23 salaried scientists (UK), their postdocs and students from seven universities, are part of the UKQCD. The costs assumed a joint Infrastructure Fund Award, which was equipped with 6.6 million pounds. Staff costs, including system support, the employed programming physicists and postdocs were approximately 1 million pounds per year, other computing and operating costs amounted to about 0.2 million pounds,
QCDOC replaces an earlier project QCDSP that the computing efficiency produced thereby, that a large number of signal processors are connected together in a similar manner.
The QCDSP computer connected 12,288 nodes to a four-dimensional network, which launched in 1998 reached 1 Tflops.
QCDOC can be viewed as precursors to the very successful IBM supercomputer Blue Gene / L ( BG / L). Both computers have a lot in common, beyond random matches. The ' Blue Gene ' is also a massively parallel supercomputer, which is composed of a plurality cheaper and relatively easier ' 440' PowerPC processors ( system on a chip, SoC). Here are these " node computer " to a high-dimensional network with a high "bandwidth" connected together. The computer, however, differ in that the processors of BG / L are more powerful and that they are connected to a faster and more effective network, comprising several hundred thousand nodes.
The compute nodes are purpose-built ASICs, each with approximately 50 million transistors. They are manufactured by IBM itself and operate at about 500 MHz PowerPC 440 processor cores. Each node has one DIMM socket, which operates between 128 and 2048 MB at 333 MHz.
The nodes operate with a total of up to 1 GFLOPS double precision.
The nodes of computers can be accommodated on a second computer dual card, with a DIMM sockets and a 4-1 Ethernet node for communication processes with other nodes. The double cards have two ports, one for the connections between the cards and one for the power supply, the Ethernet, the clock and other necessary things. 32 such computers dual card, in two rows, housed on a motherboard that supports 800 Mbit / s Fast Ethernet connection. The eight motherboards are housed in a " compartment " ( a so-called "box" ); each " compartment " contains 512 processor nodes and a total interconnect network according to a six-dimensional cube with 26 vertices. A node computer consumes a power of about 5 watts, and each " compartment " has corresponding air and water cooling necessary.
A complete system may consist of any number of " compartments ." Overall, the systems have up to tens of thousands of nodes.
Communication between the nodes
Each node can send information to the twelve nearest neighbors ( the corresponding six-dimensional lattice), and received from, with a clock rate of 500 Mbit / s Thus, a total bandwidth results of 12 Gbit / s The operating system communicates with the nodes via an Ethernet. This is also used for error diagnostics (eg communication with hard drives ) uses configuration and communication processes.
On the QCDOC a special operating system, QOS, which among other things runs the boot ( " boot " ) of the computer run time and monitoring processes enables and simplifies managing the numerous computing nodes. It uses a specific kernel and provides, inter alia, for compatibility with POSIX processes ( "unix -like" ), where it uses the Cygnus library ( newlib ).
The QCDOC is now (2010) obsolete ( see QPACE ). At the University of Regensburg, for example, he now serves as the " showpiece " in the entrance area of the physics faculty.