US chases supercomputing crown with multipetaflop Trinity system
- 11 July, 2014 06:06
Stepping up its efforts to regain supercomputing dominance from China, the U.S. within the next two years will activate what could be one of the world's fastest computers.
The U.S. Department of Energy's National Nuclear Security Administration has paid hardware company Cray US$174 million to make a multipetaflop supercomputer called Trinity that will run tests and simulations to ensure "the safety, security, and effectiveness of the U.S. nuclear deterrent without the use of underground testing," NNSA said in a statement.
Cray is using an array of new computer technologies in Trinity, which will be delivered next year or in 2016. Trinity will be located at Los Alamos National Laboratory in New Mexico.
"If this system were delivered today, it would be the fastest in the world," claimed Barry Bolding, vice president of marketing for storage at Cray.
Trinity would be the world's fastest sustained performing computer when running specific applications, Bolding said. The company couldn't provide an estimated measurement based on the Linpack benchmark currently used to measure supercomputer performance.
There's more than a year to go before Trinity will be delivered to NNSA, but it could still be among the world's fastest at that time.
The world's fastest computer is Tianhe-2 at China's National University of Defense Technology, which delivers 33.86 petaflops of peak performance, according to a Top500 list released last month. The second fastest is the U.S. Department of Energy's Titan at the Oak Ridge National Laboratory, which was made by Cray and peaked at 17.59 petaflops.
The Linux-based Trinity supercomputer will have a newer processor, memory and storage than Titan. Cray's latest Aries interconnect will contribute to the speed boost, connecting server closets, processors, storage arrays and other components. The supercomputer is a version of Cray's XC30 supercomputer, which is also sold to companies.
Trinity will have 82 petabytes of distributed storage, some of the highest capacity Cray has put into a supercomputer. It will provide throughput of 1.7Tbps (terabytes per second) for internal data transfers, faster than in the Titan system, which has Cray's older interconnect. The supercomputer will use the Lustre file system.
Trinity will also have memory and processor technologies never used in supercomputers before, which could facilitate faster performance and throughput. It will have Intel's Xeon Phi processors code-named Knights Landing, which can deliver 3 teraflops of peak performance, making it Intel's single fastest chip to date. Knights Landing is based on Intel's Silvermont CPU architecture, which is the basis for the chip maker's latest smartphone and tablet chips.
The Knights Landing chipset will have Micron's Hybrid Memory Cube technology, which provides speed and power efficiency upgrades over DDR memory. HMC provides 15 times more bandwidth than DDR3 DRAM and draws 70 percent less energy, with five times more bandwidth than the emerging DDR4 memory, which is not yet used in computers.
The Knights Landing chipset will also have DDR4 controllers, but Cray could not say whether the supercomputer would use that memory type.