The Department of Energy's Oak Ridge National Labs today powered up Titan, a new supercomputer with 299,008 CPU cores, 18,688 GPUs, and more than 700 terabytes of memory. Titan is capable of a peak speed of 27 quadrillion calculations per second (petaflops)—ten times the processing power of its predecessor at Oak Ridge—and will likely unseat DOE's Sequoia supercomputer (an IBM BlueGene/Q system at Lawrence Livermore National Laboratory) as the fastest in the world.
Based on the Cray XK7 system, Titan consists of 18,688 computing nodes, each with an AMD Opteron 6274 processor and an NVIDIA Tesla K20 GPU accelerator. The NVIDIA GPUs provide most of the computing horsepower for simulations, with the Opteron cores managing them. True to its name, Titan is big—it takes up 4,352 square feet of floorspace in ORNL's National Center for Computational Sciences.
The combination of GPUs and CPUs dramatically reduces the electrical power consumption required to generate the computing power required. "Combining GPUs and CPUs in a single system requires less power than CPUs alone."
All that effort to run a few vending machines?
Apparently they're trying to make it a bit smaller so it can fit in the next Android phone.
