With the initial port of the MWA pipeline complete, I thought it would be a good time to reflect on some of the lessons learned. Particularly relevant have been the experiences in trying to maintain a CPU and GPU based version of the code, and managing the data when everything is located in two places.

Matrix-matrix multiplications are common in quantum chemistry calculations, and can benefit enormously from GPU acceleration. Although NVIDIA provides an implementation of the BLAS *GEMM routines with its CUDA distribution, two key problems exist when trying to use these from existing code

The following news was released April 2, 2009 by NVIDIA Corporation. The company's release can be found here.

The work of the SciGPU group will move into high gear in summer 2009 as the collaborators welcome students coming to Harvard for NSF-funded Research Experiences for Undergraduates: Dominik Gothe (Univ. South Carolina; astronomy), Matthias Lee (Wentworth Inst.), Beatrice Perez (Univ. Puerto Rico; quantum chemistry), and Bo Wang (Univ. Pittsburgh; neuroscience).

The Murchison Wide-field Array is a next-generation radio telescope being built in Australia to study the early universe, the sun, space weather, and time variability of the radio sky. Cosmologists will use the MWA to map matter in the Universe during the Epoch of Reionization soon after the Big Bang, when the earliest stars, galaxies, and quasars formed. That is the MWA’s job by night, when the environment is most radio quiet.

The SciGPU project is funded through an National Science Foundation CDI (Cyber-enabled Discovery and Innovation) grant. The principal investigators on the three initial projects (the MWA telescope, quantum chemistry and the Connectome) recognized their common computational needs, and wrote a joint grant proposal. In addition to funding new researchers and hardware purchases, their proposal included a commitment to reach out to a wider community through the SciGPU.org website.

Most modern computers contain a parallel processing device of almost unprecedented power. With hundreds of processors linked to high bandwidth memory banks, this computational monster can chomp through tasks with ease. However, this device is disguised as a graphics card, and so few are aware of its existence. Unless used for playing games, the graphics processing unit (GPU) in a modern machine is little used.

Although NVIDIA recently introduced their G200 multiprocessor, most CUDA capable GPUs currently in use only have single precision arithmetic available. Some scientific calculations require double precision arithmetic for accuracy, limiting the calculations which can be put into distributed clients. We wrote MGEMM to alleviate this problem.