The U.S. Department of Energy’s (DOE) Oak Ridge National Laboratory (ORNL) unveiled their new flagship computer, Titan, on Monday. The Department also announced its latest round of Innovative and Novel Computational Impact on Theory and Experiment (INCITE) award recipients.
Titan, according to Oak Ridge’s announcement, is 10 times more powerful than its predecessor, Jaguar, with a theoretical peak performance of 20 petaflops, or 20,000 trillion calculations per second. The current fastest computer, according to the Top 500 list, is Sequoia, which clocked in at 16.32 petaflops in June.
The first phase of Titan’s installation was completed earlier this year, and final updates were completed this fall. Titan consumes slightly more energy than Jaguar, but when its significantly faster processing speed is taken into account, it is five times more energy efficient, according to the national laboratory’s team. The combination of faster speed and only slightly more energy consumption is critical, since the roughly seven megawatts Jaguar consumed — enough to power roughly 7,000 homes — cost millions. Titan is estimated to consume roughly nine megawatts.
“This power problem is changing everything,” said Steve Scott, chief technology officer of NVIDIA’s Tesla business unit. “The fact that the energy isn’t dropping as fast as the transistor budget is increasing is just making us more and more power-constrained. And that’s really what’s driving us to reinvent how we make processors.”
In addition to being faster and more efficient, Titan is the same size as its predecessor. Titan, like Jaguar, occupies a space roughly the size of a basketball court, with each stack approximately the size of a household kitchen refrigerator. That’s due to the nature of the upgrade, which primarily involved the incorporation of graphic processing unit (GPU) accelerators. GPUs are primarily used for computer games, but can be used to accelerate central processing units, or CPUs.
Titan is a Cray-XK7 system and is the first machine to use NVIDIA’s latest GPU accelerator, the Tesla K20, with each of Titan’s 18,688 nodes holding one CPU and GPU accelerator, according to Oak Ridge and NVIDIA. The GPU used in Titan is no different, said Scott, than the one made for high-end gaming units.
“The technology for gaming is the disruptive technology that’s now impacting computing broadly,” said Jeff Nichols, associate director of Oak Ridge National Laboratory.
Pairing GPUs and CPUs in and of itself is not novel, but “there were a lot of skeptics” at Oak Ridge, said Scott. It had never been done on this scale before, and Titan had to be more than, in Scott’s words, a “stunt.” The machine had to be able to run six predetermined applications to Oak Ridge’s specifications. The programs are in the areas of material science, climate change, biofuels, astrophysics, combustion and nuclear energy.
Discovers using Titan could have an impact by leading to cleaner, more efficient engines, faster and cheaper drug testing, climate modeling and even the development of future high-performance computers.
“This opens up new vistas of calculations we couldn’t conceive of doing before,” said Jeremy Smith, Governor’s Chair at the University of Tennessee and also director of the Center for Molecular Biophysics at Oak Ridge National Laboratory. Smith is likely to be one of the most frequent users of Titan, as he was with Jaguar. But Smith emphasized that Titan is merely one step in high-performance computer evolution. Smith is among many who await the arrival of exascale computing, which, in theory, would allow for, among other things, the simulation of a living cell in atomic detail.
“It’s really what comes afterwards that will provide the bulk of the discoveries,” said Smith. “What Titan will have done is to set the standard in computer power, identify the challenges in using such a machine, and it will make a couple of useful discoveries that couldn’t be made on any other machine.”
Smith will not be alone in leveraging Titan’s processing power. Recipients of the 2013 INCITE awards will also have access to Titan. The Department of Energy’s Leadership Computing Facilities (LCFs) awarded a total of 4.7 billion hours to 61 projects in science and engineering — 1.84 billion hours on Titan and 2.83 billion hours on two of Argonne National Laboratory’s supercomputers, Mira and Intrepid. Projects ranged from research around nuclear reactors and electric engines to the development of a unified theory for physical forces.
Titan’s public unveiling comes weeks before the release of the latest Top500 supercomputer rankings. The Top500 list, which dates back to 1993, is released twice a year — once in June and again in November. The Titan team anticipates their machine will be ranked in the top two, which would make it the fastest high-performance computer open to non-classified projects. That’s assuming Sequoia comes in first or second. Sequoia is housed at Lawrence Livermore National Laboratory (LLNL) and is used exclusively by the National Nuclear Security Administration (NNSA) to manage the United States’ nuclear weapons stockpile. Jaguar ranked sixth in the latest list.
When asked what he would use Titan for, Nichols said he thought there was “fascinating” research to be done in chemical physics, specifically simulating the breaking of chemical bonds. But given his focus in materials science, he said he would most likely use Titan to figure out how to design better photovoltaics.
NVIDIA’s Scott, on the other hand, said he would use the supercomputer to help him find Titan’s successor. But a few moments later, added, “You could also maybe use the machine for a really giant multi-user game.”
Read more news and ideas on Innovations: