PORTLAND, Ore. -- The fastest supercomputer in England is the recently installed Academic Research Computing High End Resource (Archer) at the science and engineering oriented University of Edinburgh. Performing over 1.3 million billion (peta) operations per second (OPS), Archer is almost four times faster than the previous High-End Computing Terascale Resource (Hector) supercomputer it replaces.
"Archer is roughly 70,000 times as powerful as a standard laptop. Tasks that would take a year to run on your laptop can be run in a matter of minutes," said computational science and engineering director Alan Simpson in a video interview:
Funded by UK's Engineering and Physical Sciences Research Council (EPSRC), the $60 million Archer system will provide high-performance computing support for research and industry projects in the England, including modeling the Earth’s climate, simulating the airflow around aircraft, and designing new materials.
"Some of the new applications we are seeing a very interesting, because they are much more able to be tied to the economic impact on the man in the street -- people like my family, not just my colleagues," said Simpson. "Things like climate modeling [are] a very big issue, and understanding how quickly the climate is likely to change and how far it will go is important in understanding what our future lives will be like. On more day-to-day things, designing more efficient planes, more efficient cars… Fuel efficiency is important for everybody -- more fuel efficiency saves money and reduces pollution."
Archer is based on a Cray XC30 liquid-cooled chassis housing over 6016 Intel Xeon E5-2600v2 12 core processors running at 2.7-GHz and connected by an Aries interconnect. The total of 76,192 dual-thread cores -- with 144,384 total threads -- are divided in to groups of 24 cores per node. Seven groups of 376 standard nodes share 64-Gigabytes of memory among their 24 cores per node, for a total of 2632 standard nodes with 63,168 cores. One group of 376 high-memory nodes, with 9,024 cores, shares 128 Gigabytes of memory among its 24 cores per node. In all, Archer houses 3008 compute nodes each, with 24 cores divided into eight groups, with 72,192 cores total.
Cray XC30 series chassis are based on blades with processor daughter cards (PDCs) and a peripheral component interconnect express (PCIe) Gen3 host interface.
When idle, Archer draws 400 kWatts of power. At peak performance, with all cores running full speed, it draws 1,200 kWatts of power. Archer uses the Dragonfly interconnection topology, consisting of one Aries routing and communications application-specific integrated circuit (ASIC) per four computing nodes, each with a total of 48 switch ports running at 500 Gigabytes per second. Each set of 188 nodes are grouped into a single cabinet, for 16 cabinets total.
Cabinets are grouped into pairs with two-dimensional all-to-all copper interconnects between all nodes in a group, and all-to-all optical interconnects between groups. The number of optical interconnects between groups can be varied according to system requirements, with a total of 84 optical links per group, for a peak bi-section bandwidth of 7,200 Gigabytes per second for the whole system. The message passing interface (MPI) latency is 1.3 microseconds over Aries, with an addition 100 nanoseconds of latency when using an optical link.
The University of Edinburgh recently installed its Archer supercomputer, which ranks as the fastest supercomputer in England.
(Source: University of Edinburgh)
Archer is housed at the University’s Advanced Computing Facility at Easter Bush. It runs the Cray Linux Environment operating system. Archer is supported by the newly installed UK Research Data Facility, which was created to support Big Data applications identified by the UK government as one of its Eight Great Technologies.
— R. Colin Johnson, Advanced Technology Editor, EE Times