TACC engineers by 2018 will replace the current Stampede supercomputer with new Dell servers running the latest Intel Xeon and Xeon Phi chips.
AUSTIN, Texas—The Dell-based Stampede supercomputer at the Texas Advanced Computing Center here, which currently is the 10th
-fastest system in the world, is about to get a significant upgrade.
Officials with the center, which is part of the University of Texas at Austin, announced June 2 that the facility will use a $30 million award from the National Science Foundation (NSF) to build a new large-scale supercomputer that will provide twice the peak performance, memory, storage capacity and bandwidth of the current system, which went online in 2013.
Stampede 2 will be built with Dell PowerEdge servers and powered by a mix of the next generation of Intel Xeon chips and the chip maker's many-core Xeon Phi "Knights Landing" processors. It also will use Intel's OmniPath connectivity fabric and the chip maker's upcoming 3D XPoint memory technology. Once it's all in place in 2018, the supercomputer will deliver 18 petaflops of peak performance, twice the amount of the current Stampede. Eighteen petaflops would put the system in the No. 2 slot on the most recent Top500 list
of the world's fastest supercomputers, which was released in November 2015.
The next list will be released later this month at the International Supercomputing 2016 show in Germany.
The announcement was made June 2 during a celebration of the 15th
anniversary of the Texas Advanced Computing Center (TACC) and the opening of a new section of the center that will house offices, an auditorium, a training center and other functions.
Speaking to a group of analysts and journalists here the day before, TACC Executive Director Dan Stanzione said the new, larger supercomputer will provide high-performance computing (HPC) capabilities to thousands of researchers across the United States, meeting what he said is a fast-growing demand for access to the system. He noted that the original system has run more than 7 million simulation and data analysis tasks for tens of thousands of researchers from the United States and around the world.
Engineers at TACC will be tasked with phasing in the new supercomputer while the original Stampede continues to run. They will remove parts of the current system at a time and replace them with the new technologies. By 2018, the bulk of the supercomputer will be new.
"Eventually, we will replace every part of the system," Stanzione said.
TACC has already installed 500 Knights Landing nodes as part of an upgrade in what Stanzione called Stampede 1.5. Intel unveiled the 14-nanometer Knights Landing processor
at the SC 15 supercomputer show in November 2015. It's the latest version of the many-core chips, which first rolled out in 2011. The original Xeon Phi was a 22nanometer 60-core chip designed to be used as a coprocessor that works alongside primary Xeon processors, playing a role similar to the GPU accelerators offered by Nvidia and Advanced Micro Devices.
With Knights Landing, Intel is offering a 14nm chip with as many as 72 cores and which can be used as either a coprocessor or primary processor. Built on the company's Silvermont architecture, Knights Landing has more than 8 billion transistors and will deliver more than 3 teraflops of peak performance, Intel officials have said.
The first Knights Landing chips used in Stampede 2 will not include the integrated OmniPath interconnect; instead they will use cards. Chips with the integrated fabric will come later starting in 2017. Many of them will be used as primary chips rather than coprocessors, according to Stanzione.
Stampede—and eventually Stampede 2—is only part of TACC. The center, which has a staff of 135, has three data centers totaling 12 megawatts of capacity housed in two buildings that run more than 15 supercomputers. In total, the center runs about a billion compute hours a year and 50 to 60 petabytes of data. The bulk of the supercomputers run Dell systems, though there are some that are powered by other infrastructure products from the likes of Cray and Hewlett Packard Enterprise, including its Apollo 8000 servers.