More supercomputer news this week: The US is responding to China’s new Sunway TiahuLight system that was announced Monday, and fast. First, the Department of Energy’s (DOE) Oak Ridge National Laboratory is expected to take delivery of a new IBM system, named Summit, in early 2018 that will now be capable of 200 peak petaflops, Computerworld reports. That would make it almost twice as fast as TaihuLight if the claim proves true. (We had originally reported in 2014 that both Summit and Sierra would achieve roughly 150 petaflops).
TaihuLight (pictured below) now sits at number one on the twice-yearly TOP500 list of the fastest supercomputers in the world, with a Linpack benchmark score of 93 petaflops and a claimed peak of 124.5 petaflops. The latest TOP500 announcement Monday caused a bit of a stir. Not only is TaihuLight roughly three times faster than China’s Tianhe-2, the prior champion, but it also uses no US-sourced parts at all for the first time, as it’s powered by Sunway 260-core SW26010 processors that are roughly on par with Intel Xeon Phi, as well as custom proprietary interconnect.
In turn, Summit will employ IBM Power9 and Nvidia Volta GPUs. Summit will deliver over five times the computational performance of Titan’s 18,688 nodes using only about 3,400 nodes. Each node will have “over half a terabyte” of so-called coherent memory (HBM + DDR4), plus 800GB of non-volatile RAM that serves as a burst buffer or extended memory.
Titan (pictured top), meanwhile, is currently #3 on the TOP500 list, and resides at the DOE Oak Ridge National Laboratory, where the new IBM Summit system will be located. In a statement via Computerworld, the DOE pointed out that since 1993, US supercomputing capabilities “have grown exponentially by a factor of 300,000,” and that “high-performance computing remains an integral priority for the DOE.” (For more on Oak Ridge’s earth, energy, climate, and geographic science research using these systems, visit its dedicated supercomputing page.
The Linpack benchmark has emerged as a singular yardstick for measuring the performance of complex supercomputers. It doesn’t record overall performance in all situations; rather, it measures the performance of a system when solving a “dense system of linear equations” that gives a good approximation of real-world (as opposed to peak) performance. Read more about the benchmark at the above link.
Separately, Cray announced this week at the 2016 International Supercomputing Conference in Frankfurt, Germany that its Cray XC systems are now available with the latest Intel Xeon Phi (Knights Landing) processors. The company said the new XC systems, which feature an adaptive design that supports multiple processor and storage technologies in the same architecture, deliver a 100% performance boost over prior generations. Cray also unveiled the Sonexion 3000 Lustre storage system, which can deliver speeds of almost 100GB/sec in a single rack.