The Dept. of Energy keeps adding number-crunching power in the form of supercomputers to its arsenal of scientific and engineering tools. At the Oak Ridge National Laboratory in Tennessee, for example, the new Titan supercomputer replaced the XT5 Jaguar as the world’s fastest computer on the Top500 list, a semiannual ranking of the world’s supercomputers. Titan also placed third on the Green500 list, a project at Virginia Tech that ranks computers by the number of calculations per watt of electricity. Titan is a Cray XK7 with 18,688 nodes, each built from a 16-core AMD Opteron 6274 processor and a NVIDIA Tesla K20X GPU accelerator. It has 710ŒTbytes of memory and a theoretical top speed of 27 petroflops or 27Œquadrillion calculations per second. Running at top speed takes about 9 MW of electricity, enough to power 9,000 homes.
Titan’s top speed makes it 10 times faster than Jaguar but it uses only 20% more electrical power, thanks to efficiency gains from using GPUs. GPUs, or graphical-processing units, were developed for the computer gaming industry. But because they can handle hundreds of calculations simultaneously, GPUs can perform more calculations than CPUs in a given time. So Titan relies on its 299,088 CPU cores to guide simulations while the NVIDIA GPUs do the heavy lifting, which leads to faster, more-accurate calculations and less power consumption. In one task, Titan took 13 hr to simulate the life of a fuel rod through one round of use in a reactor, a job that took the Jaguar XT5 60 hr.
The Energy Dept. is also expecting to accept delivery of a new supercomputer at the Environmental Molecular Science Laboratory at its Pacific Northwest National Lab. Built by Atipa Technologies, Lawrence, Kans., the new computer will be used to research climate and environmental science and explore biology-based fuels. The $17 million machine will have a peak speed of 3.4 petaflops and will do more calculations in an hour than a laptop could do in 20 years.
Its 196,000 processing units include Intel processors and MIC accelerator cards. The computer will also have 184,000 gigabytes of memory along with 2.6 petabytes of storage, about four times more memory per processor as other supercomputers. The additional memory is said to make the machine more efficient.