QCDOC

QCDOC
Name	QCDOC
Developer	IBM, Columbia University, RIKEN
Release	2000s
Type	Supercomputer
Cpu	PowerPC-based ASIC
Memory	DDR SDRAM per node
Os	Custom microkernel, Linux variants
Purpose	Lattice gauge theory, computational physics

QCDOC QCDOC was a family of purpose-built supercomputers designed for lattice quantum chromodynamics calculations, developed through a collaboration among IBM, Columbia University, and RIKEN. The project combined custom hardware, dedicated interconnects, and tailored software to deliver large-scale simulations used by researchers at institutions including Brookhaven National Laboratory, Lawrence Berkeley National Laboratory, and University of Edinburgh. Its design emphasized low-latency message passing, energy efficiency, and scalability for sustained floating-point performance in production physics runs.

Overview

QCDOC targeted high-volume lattice field theory workloads for groups such as Fermilab, CERN, SLAC National Accelerator Laboratory, and University of Illinois at Urbana–Champaign. The machine used a custom application-specific integrated circuit developed by teams from IBM Research and academic partners, drawing on architectures similar to those in Blue Gene. QCDOC installations were managed by centers including Riken BNL Research Center and supported experimental programs at Thomas Jefferson National Accelerator Facility, Oak Ridge National Laboratory, and MIT. Funding and program oversight involved agencies such as the Department of Energy, the National Science Foundation, and Japan’s Ministry of Education, Culture, Sports, Science and Technology.

History and Development

Initial conception began in collaborations among researchers at Columbia University, Brookhaven National Laboratory, and IBM Watson Research Center. Early design meetings referenced experience from projects like ASCI Red, Blue Horizon, and discussions at conferences such as the International Conference on High Performance Computing, Networking, Storage and Analysis and workshops at MIT. Prototype development involved testbeds at University of Edinburgh and production planning included participation by RIKEN and Brookhaven National Laboratory. Deployment phases coincided with contemporaneous architectures like QCDOC-era machines from Cray Research and lessons from Intel Paragon and Fujitsu systems.

Architecture and Hardware

The QCDOC node integrated a PowerPC core with a custom communication unit on a single ASIC designed by IBM Research engineers who previously worked on POWER designs. Nodes included local DDR memory managed by controllers implemented on-chip and utilized a six-dimensional torus network topology resembling topologies used by Blue Gene/L and other large-scale systems at Lawrence Livermore National Laboratory. Chassis and cabinets were manufactured in collaboration with firms linked to IBM, deployed in facilities at Brookhaven National Laboratory, RIKEN, and university computing centers such as University of Edinburgh and Columbia University. Cooling systems and power provisioning followed standards set by data centers at Oak Ridge National Laboratory and Argonne National Laboratory.

Software and Programming Model

QCDOC systems ran a lightweight microkernel tailored for low-latency communication, with higher-level support from Linux variants and runtime libraries developed by teams at Columbia University and IBM Research. The programming model emphasized message-passing interfaces compatible with MPI implementations used at NERSC and custom communication libraries optimized for the torus interconnect. Compilers and tools included GNU toolchain ports, IBM compilers influenced by work for Blue Gene/L, and performance analysis tools similar to ones used at Lawrence Berkeley National Laboratory and National Center for Supercomputing Applications. Users from projects at Fermilab, CERN, Riken BNL Research Center, and Brookhaven National Laboratory wrote applications in C, Fortran, and assembly to exploit the architecture’s SIMD and floating-point units.

Performance and Applications

QCDOC delivered high sustained performance on lattice quantum chromodynamics kernels, enabling simulations for collaborations involving MILC, UKQCD, and groups at Brookhaven National Laboratory and Columbia University. Benchmarking compared favorably with contemporary systems such as Cray XT3 and early Blue Gene machines for targeted workloads, showing strong scalability for problems studied by Particle Data Group researchers and theorists at CERN and KEK. Scientific applications included hadron spectrum calculations, thermodynamics studies pursued by teams at Ohio State University and University of Tokyo, and algorithm development in concert with researchers at MIT and University of Edinburgh. Performance tuning efforts referenced methodologies from Top500 submissions and profiling techniques used by centers like NERSC.

Deployment and Legacy

QCDOC installations appeared at research centers including Brookhaven National Laboratory, RIKEN, Columbia University, and several university clusters coordinated with agencies such as the Department of Energy and National Science Foundation. The project influenced later designs, informing architectures like Blue Gene/L and contributing to expertise at IBM and academic labs involved in exascale planning at Oak Ridge National Laboratory and Argonne National Laboratory. Software artifacts and lessons from QCDOC seeded community codes maintained by collaborations such as MILC and influenced programming models adopted by groups at CERN, Fermilab, and Lawrence Berkeley National Laboratory. Its legacy persists in specialized hardware approaches for computational physics and in training of researchers who later led projects at NVIDIA, Intel, and national computing initiatives.

Category:Supercomputers