A CT scan produced this volume rendering of a mini-flashlight

Soon, There's Roadrunner—€“It's Really Really Fast

While it's hard to believe that your 14-year-old son, your daughter's 23-old boyfriend, or the 12-year-old next door has anything to do with Los Alamos National Laboratory's newest supercomputer—€”you might be surprised to learn they have had a lot more influence than you think.

Thanks to millions of video gamers who demand that their games be as realistic as possible, microprocessor companies have had to step up their game to meet the demand by incorporating faster processors into their products. The Sony Playstation 3 game console has a processor developed by IBM, Sony and Toshiba called Cell Broadband Engine—„, or Cell B.E. Cell B.E. is the key to the real-life quality of graphics and animation in the Playstation 3 as well as the key to Los Alamos' Roadrunner supercomputer, which is scheduled to be up and running in 2008.

"For the past 60 years—€”almost since the founding of the lab in 1943—€” computer simulations of physical and biological systems have played an important role in the science done at the laboratory," said Nehal N. Desai, one of many scientists working with Roadrunner. "As the sizes of the simulations have grown, so have the computational and data requirements of the computer systems. In order to satiate this appetite for computational power the laboratory has been a proponent of supercomputers. Beginning with MANIAC in the 1950s and continuing today with the ASCI [Advanced Scientific Computing Initiative] Q machines, Los Alamos has been a leader in the use of novel high performance computer systems for scientific calculations."

Additionally, industry leaders such as General Motors, Procter & Gamble and many others are leading the growing trend of using supercomputers for their business processes.

In 2006, the Department of Energy's National Nuclear Security Administration selected Los Alamos as the development site for Roadrunner and IBM as its designer and builder. Roadrunner, which is expected to cost $110 million, is known as the first "hybrid" supercomputer because of its unique architecture.

A typical supercomputer is constructed with a cluster of computers each having a certain number of central processing units, or CPUs. For example, Blue Gene/L at Lawrence Livermore National Laboratory, currently the fastest supercomputer in the world, is made up of over 65,536 computer nodes, each with two CPUs. LINUX-based Roadrunner, instead, has more than 16,000 CPUs, Advanced Micro Devices's Opteron processors and more than 16,000 Cell B.E. processors.

Cell B.E. is what really makes Roadrunner unique. Unlike the CPU, often referred to as a core, found in most desktop computers, Cell B.E. is known as a heterogeneous multicore processor. Cell B.E. is made of nine cores that together are a formidable computational engine and consume less energy than a CPU. One Cell B.E. can, on certain applications, deliver an order of magnitude performance improvement over most CPUs.

Because Cell B.E. processors are 10 to 15 times faster than CPUs, the most complex and repetitive parts of an application will be directed to the Cell B.E. processors. These parts of the application are the largest consumers of computing power during a calculation; by directing them to the Cell B.E. processors, Roadrunner will be able to run applications faster than any other supercomputer in the world.

With Cell B.E. processors pushing calculation speeds, Roadrunner is poised to reach speeds of more than 1,000 trillion operations per second, or one petaflop. When Roadrunner reaches this milestone, it will be almost four times faster than Blue Gene/L.

While Roadrunner is ideal for the colossal calculations and simulations Los Alamos scientists would like to run faster, there is one challenge that must be overcome—€”the programming language. The CPUs handle data much differently than Cell B.E. and even graphics accelerator cards. Graphics accelerator cards are similarly constructed to Cell B.E. but used specifically to handle graphics, each one usually having between 64 and 128 cores. While CPUs understand more complex code, Cell B.E. and graphics accelerator cards understand only highly specialized data. Los Alamos scientists, who must start almost from scratch, are just getting started on code for Roadrunner's unique integrated system.

Los Alamos computational scientist Patrick McCormick and his team have been working on developing code for graphics accelerator cards for nearly five years. Graphics accelerator cards, although used specifically for graphics on most computers, are capable of both basic computation and visualization analysis and are somewhat easier to write code for than Cell B.E. McCormick's original goal was to make larger calculations possible on a desktop computer. This way scientists would be able to have semi-supercomputing capabilities at an affordable price (graphics accelerators are much cheaper than accelerators like Cell B.E.). The result of McCormick's work is a product dubbed Scout.

Scout is a programming language that allows scientists to use the graphics accelerator to do general purpose computations, visualization and data analysis.
Because Scout has proven to be a valuable tool for graphics accelerators, McCormick and his team will attempt to use what they've learned from their experience to write programming language for Roadrunner and its Cell B.E. processors.

"Our next goal with Scout is to allow scientists to use it to program hybrid applications," said McCormick. "This is very much a research effort that is really just getting started. The ability to program these multi-core, or many-core (the new buzzword), processors is a challenge to the entire computing industry."

Once Roadrunner has a solid programming environment, it will be of great value to the Los Alamos scientific community. Additionally, supercomputers are beginning to appear all over from use in the entertainment industry for animations and graphic simulations, to auto accident simulations for safety ratings and the search for oil and gas deposits, among a multitude of uses. All of these uses will save companies plenty of time (because the simulations take hours opposed to days, or days opposed to weeks of research) and money.

Steve Stringer, an industrial fellow with the lab's technology transfer division, is the liaison between the laboratory and Procter & Gamble, where he works to connect the company's business needs with Los Alamos's scientific capabilities.

"It's well known that P&G is interested in connecting with many outside sources to help develop its products and processes," Stringer said. "We share an interest with P&G in developing advanced applications (such as molecular dynamics) that are well-suited for running on Roadrunner's hybrid architecture.
While our end products and our customers are dissimilar, the first principles underlying chemistry, physics, mechanical engineering, materials design and life sciences are exactly the same for us as they are for them."

It's anticipated that other private-sector companies will make use of Roadrunner, with laboratory input. Hybrid architecture seems to be the next step toward making personal computers orders of magnitude faster. Already, desktop computers are using dual processors and, like Roadrunner, hybrids look like the next option. The Scout code and McCormick's new mission to write programming language for Roadrunner may be a first step toward writing code for future desktop and laptop computers.

Krystal Zaragoza is a communications specialist at Los Alamos National Laboratory.