QCDOC:
QUANTUM CHROMODYNAMICS
ON A CHIP

2005–2011

EPCCsocial
@EPCCsocial

QCDOC was a massively parallel, custom-built machine. It consisted of low-power processing elements (500 MHz 440 PowerPC).

It was capable of focusing significant computing resources on relatively small but extremely demanding problems in the field of quantum physics.

QCDOC was designed and jointly built by University of Edinburgh (part of the UKQCD consortium), Columbia University, RIKEN BNL, Brookhaven Research Centre and IBM.

 There were three 10-Tflops machines in service: one hosted by EPCC and two in the US.

The system WAS created TO SIMULATE LATTICE QUANTUM CHROMODYNAMICS. 

In particle physics, CP (charge parity) violation is thought to be crucial in explaining why matter dominates over antimatter in the Universe and in the study of weak interactions in particle physics, which are the only type of known interactions breaking CP symmetry. 

One of the processes that plays a central role in constraining the structure of the Standard Model (a theory of fundamental particles and how they interact) in the search for new physics is indirect CP violation in Kaon decays. Neutral Kaons can transform into their antiparticles and vice versa, however, this transformation does not occur with exactly the same probability in both directions. 

The simulations that were run on QCDOC systems as a part of Riken-BNL-Columbia/UKQCD collaboration on the Kaon/anti-Kaon mixing parameter led to a reduction in the uncertainty in this parameter from around 15% to around 3.5%. Although we still do not know whether the Standard Model correctly predicts the CP violation in Kaons, reducing the uncertainty increases that probability. 

The work to further improve the precision of Kaon mixing parameter (to around 0.5%) was performed on Blue Gene/Q.

Research: Quantum fluctuations in chiral fermion fields associated with the Feynman path integral changing the topology of the gluon field.

These spikes are created by the famous Atiyah Singer index theorem for which the mathematician Sir Michael Atiyah won the 1966 Fields medal and the 2004 Abel Prize. Atiyah has been an honorary Professor at Edinburgh since 1997

About the System

CRAIG MORRIS, EPCC

QCDOC was a supercomputer technology focusing on using relatively cheap low-power processing elements to produce a massively parallel machine. 

As the name suggests, the machine was custom-made to solve small but extremely demanding problems in quantum physics. While based at the University of Edinburgh, QCDOC had 12,288 processors and a peak speed of 10 teraflops. To give an idea of the complexity and the amount of processors and networking on this machine back in 2004, it had 192 motherboards with 12,288 processors, 2304 serial communication cables, 1608 ethernet cables and 78 network switches. The main advantage of the complexity of this machine was its scalability: it could employ between 2 and 12,288 processors, if required. 

The system was used by the QCD community, and EPCC hosted the service and maintained it for around 6 years.

Each of QCDOC's nodes comprised a single applications-specific integrated circuit (ASIC) chip containing 500 MHz 440 PowerPC processor core with 4 Mbytes embedded DRAM (EDRAM). 

This on-chip memory was sufficient to fit the code and data for standard lattice QCD calculations.

The ASIC had direct memory access (DMA) capability for moving data between EDRAM and external memory, and circuitry to support communication between nodes. 

QCDOC ran QOC, a custom-built operating system facilitating boot, runtime, monitoring, diagnostic and performance checks. The QOC also maintained system partitions allowing different applications to run on different partitions but with only one client application per partition at any given time. 

QCDOC can be seen as a predecessor to Blue Gene/L, which had more powerful computing nodes connected by faster and more complex network, allowing the system to scale up to several hundred thousand nodes.