Nvidia has finished work on a new supercomputer at Oak Ridge that showcases the company's K20 Tesla
GPUs. The new cluster, dubbed Titan, is expected to break the 20 petaflop barrier. The system is a major upgrade to the Jaguar supercomputer that proceeded it (Jaguar was prominently upgraded last year to AMD's new Bulldozer
CPUs after utilizing Magny-Cours products. The net effect of the upgrade has boosted Titan's raw computing power tenfold, while occupying the same floor space as the previous Jaguar system.
The supercomputer now contains 18,688 nodes (299,008 CPU cores), with each node built around the AMD Opteron 6274 and Tesla K20. "One challenge in supercomputers today is power consumption," said Jeff Nichols, associate laboratory director for computing and computational sciences. "Combining GPUs and CPUs in a single system requires less power than CPUs alone and is a responsible move toward lowering our carbon footprint. Titan will provide unprecedented computing power for research in energy, climate change, materials and other disciplines to enable scientific leadership."
The system will be deployed on projects designed to investigate materials science at a nanoscale level, model internal combustion at greater levels of detail to improve engine efficiency, improve climate change research, and measure the behavior of neutrons in aging nuclear reactors.
Getting the machine up and running is itself a significant achievement. The challenge of exascale computing is an enormous one and DARPA
challenged manufacturers to submit solutions that would allow supercomputers
to perform at up to one exaflop within a reasonable time frame. The K20 is still a long, long way from that goal, but Nvidia's power efficiency gains should translate into valuable progress in the right direction.
“Science and technology have always been our primary goal, and Titan is a groundbreaking tool that will allow researchers worldwide to leverage GPU
-accelerated computing to make unparalleled breakthroughs,” said Jeff Nichols, associate laboratory director for computing and computational sciences at Oak Ridge National Laboratory. “The new Tesla GPU accelerators offer the performance and energy efficiency that enable Titan to scale to unprecedented performance levels without consuming the energy equivalent of a small city.”
The loadout remains impressive, if slightly below city scale. Jaguar now contains 710TB (yes, terabytes) of total RAM across its various nodes. It's also a major win for Nvidia, which faces increased competition from Intel's Knights Corner / Xeon Phi. Which of these two solutions will take the accelerator crown is very much in question, but we've never doubted that the K20 would compete well for the winner's circle.