The Yellowstone supercomputer is the 50th fastest supercomputer in the world — but it could soon become even faster.
Thanks to a recent collaboration, the National Center for Atmospheric Research-Wyoming Supercomputing Center could be getting a replacement for the aging machine.
Rich Loft, the director of technological development in the computation and information systems lab at NCAR, said the 3-year-old computer is nearing the end of its life.
“Yellowstone, as far as supercomputers go, is nearing retirement age,” he said. “We’ll run it for another two years before we turn it off — we should have a new machine in our facility by the start of 2017.”
The new computer could top the www.top500.org list of fastest computers — a list created by “high-performance computer experts, computational scientists, manufacturers, and the Internet community,” the website states — thanks to a recent collaboration with the computer hardware giant Nvidia, said Suresh Muknahallipatna, professor of electrical and computer engineering.
“This is cutting edge,” he said. “It’s the very latest for new technology, and it’s released to us before public consumption.”
Nvidia produces graphics processing units (GPUs), widely known for their use in computer gaming, Loft said. However, in recent years, GPUs have been used increasingly in supercomputers for superior speeds in comparison to central processing units (CPUs), which supercomputers traditionally use.
The main positive is the number of cores in a GPU chip.
“A single processing unit is called a core,” Loft said. “A core would be where you put a process that is doing math.”
The Yellowstone computer currently uses CPU chips with 8 cores each. A new computer could make use of Nvidia GPU chips which each hold 1,000 cores.
“Nvidia is using the same underlying silicon chip, but they’ve made the processors much, much simpler than the ones on Yellowstone,” he said. “They’re not running as fast, but there’s more of them.”
While the large number of cores lead to faster times, it creates more work and challenges for computer engineers and developers.
“It’s a lot harder to program 1,000 things to cooperate with one another as it is with eight things,” Loft said. “Processors wander off and do some other task, and everything has to wait for it. The whole idea of programming is to eliminate this time waiting for things to catch up.”
One reason Nvidia decided to partner with the University of Wyoming is because of current research trying to find the best way to utilize GPUs, said Jeff Clune, associate professor of computer science.
“People at this university were taking advantage of (GPUs), pushing the limits,” he said. “Nvidia took notice and suggested we apply for a formal partnership.”
The partnership allows UW access to Nvidia’s latest equipment before the public can purchase them, Muknahallipatna said, while also providing support for any problems university students or professors might encounter in their research. A discount for future chips is also provided.
Early access to chips could be the most important part of the deal, Loft said, as it presents a huge opportunity to students.
“The real benefit is, this creates a place for the brightest students in the field to come and work on this particular kind of technology,” he said. “They are working on what will be the future, and I always like to see students injected into the industry.”