Parallel Computing with QCDSP and QCDOC

The QCDSP computers at Columbia University and the RIKEN-BNL Research Center total 20,000 processors and are in full-time use for lattice QCD simulations. These inexpensive computers offer a combined peak performance of 1 Teraflops and sustain around 300 Gflops for QCD. Some of the features of these systems will be discussed as well as the science they are doing.

Currently under design, the QCDOC (QCD On a Chip) computers will involve a similar number of processors, but will be based on a single custom ASIC containing a Power-PC processor and low-latency embedded DRAM. The full double precision architecture of the processor and the ability to add external DRAM make this new computer a powerful resource for QCD and other applications. Aspects of the design and use of this machine will be discussed.

Top of Page