Highlights of our Work
The most powerful compute engine in your laptop and home computer is what? The central processing unit, of course. Wrong, it is the graphics processing unit (GPU)! Since GPUs are specialized for graphics and games, popular uses of modern computers, their development benefitted from strong market forces. Modern GPUs reach Teraflop speed, passing entire computer clusters filling a room. Unfortunately, GPUs lie dormant when computers are employed for intensive biological computing like biomolecular simulations. GPUs are now getting a wake-up call from biomedical researchers who then enjoy many-fold speed-ups of their laptop computations (see the Oct 2007 highlight), vendors already now hawking GPU-powered "desktop supercomputers." The next challenge is to bring the GPU from the laptop to computer centers, speeding up the world's fastest supercomputers. Building on previous development experience of NAMD and VMD, a recent report describes NAMD running at impressive speed on a GPU cluster. The report suggests techniques, useful also for programmers of other applications to efficiently accelerate computer clusters through GPUs. The report is particularly timely with a new, large GPU-accelerated computer cluster coming on-line. The reported GPU-based speed-ups will permit biomolecular simulations largely unfeasible so far for studies of entire virus particles and cellular organelles. More information here.