Researchers at the University of Arizona now have a world-class supercomputing system to help them unlock the secrets of the universe. “El Gato” – which is short for the Extremely LarGe AdvancedTechnOlogy system – was acquired via a $1.3 million grant from the National Science Foundation. The 145-teraflops system achieved a 336 ranking on the TOP500 list and placed seventh on the Green500 list, which reshuffles the TOP500 deck based on FLOPs-per-watt.
With 13 times more processing capability than the previous generation HPC system in the UA’s Research Data Center, “El Gato” will enable the UA faculty to reach new heights when it comes to understanding complex scientific phenomena, such as the distribution of dark matter in the universe. The machine has already calculated the location of more than one billion dark matter particles in a simulation of the universe that is 280 million light years from side to side.
University of Arizona assistant professor of astronomy Brant Robertson led the committee to bring this next generation system to the UA campus. The collaborative effort involved astronomers, computer scientists and engineers. The $1.3 million grant came from the National Science Foundation (NSF) Division of Astronomical Sciences through the Major Research Instrumentation program. El Gato, constructed of Intel, IBM and NVIDIA parts, was installed in December 2013 in the UA Research Data Center, a facility that supports all UA researchers.
Like many other HPC systems being built today, El Gato draws its power from a combination of CPUs and GPUs. The hybrid architecture, which uses GPUs to accelerate workloads, is one of the biggest trends to hit HPC in the last decade.
“For the price, this computer is very, very fast, and it’s very green,” Robertson said. “The graphic processing units enable you to speed up your calculations up to 300 times faster compared with central processing units.”
It’s a degree of power that’s not to be taken lightly. “This computer is very powerful for a university of this size,” observes Robertson’s graduate student Evan Schneider. “For the University of Arizona to have a computer that is ranked in the top 500 fastest computers in the world is pretty impressive.”
As Robertson points out, the top one hundred fastest machines in the world (those on the TOP500 list and those that are not disclosed to the public) are primarilty the purview of national research facilities, government labs and similar agencies, as well as very large corporations with budgets that can accommodate the hefty price tag.
The new supercomputer will support faster compute times and much more detailed results.
“El Gato allows us to perform bigger calculations, to look at finer details, and to include more features in our models,” Robertson said. “All of this enables us to do research we couldn’t do otherwise.”
While the main area of focus will be in theoretical astrophysics, any UA researchers – including faculty and students – can access the system via the “windfall” usage policy that redirects “idle” time to worthy projects. “Anyone who is eligible for a high-performance computing account can use the system,” Robertson said. “We have reserved 30 percent of the total usage time for the UA researchers and their collaborators. Most of the time it is used by people not associated with the grant, which is exactly what we wanted.”
The hybrid CPU-GPU architecture is new to many of the system’s users, who were accustomed to the school’s more traditional CPU-based HPC resources. To address this, the team of project co-PIs along with their grad students and postdocs have been retooling codes to make optimal use of the GPU accelerators. Current projects include the direct imaging of black holes, simulating the dynamics of clusters and reproducing the formation of galaxies. Robertson hopes that El Gato will encourage users to develop GPU-optimized code for other fields too.
The computer’s suppliers – IBM, NVIDIA and Intel – will also have a hand in educating El Gato’s broad user community. This September, the companies will help lead an El Gato training symposium.