Tianhe2 then again TH2
"Heavenriver2", that is,
"Smooth Way 2") is a 33.86petaflop supercomputer situated in National
Supercomputer Center in Guangzhou. It was produced by a group of 1,300 researchers
and engineers. It was the world's quickest supercomputer as indicated by the
TOP500 records for June 2013, November 2013, June 2014, November 2014, June
2015, and November 2015. The record was outperformed in June 2016 by the Sunway
TaihuLight. In 2015, arrangements of the Sun Yatsen College as a team with
Guangzhou region and city organization to two fold its registering limits were
ceased by a US government dismissal of Intel's application for a fare permit
for the CPUs and coprocessor sheets.
In light of the US endorse, China presented
the Sunway TaihuLight supercomputer in 2016, which significantly beats the
Tianhe2, also, now holds the title as the quickest supercomputer on the planet
while utilizing totally household innovation including the Sunway multicore
microprocessor.
Historical
Development of Tianhe2:
The advancement of Tianhe2 was supported by the
863 High Technology Program, started by the Chinese government, the
administration of Guangdong region, and the legislature of Guangzhou city.
It
was worked by China's National University of Defense Technology (NUDT) as a
team with the Chinese IT firm Inspur. Inspur made the printed circuit sheets
and assisted with the establishment and testing of the framework software. The
venture was initially planned for consummation in 2015, yet was rather
announced operational in June 2013. As of June 2013, the supercomputer had yet
to end up completely operational. It was relied upon to achieve its full
registering capacities before the end of 2013. In June 2013, Tianhe2 topped the
TOP500 rundown of quickest supercomputers on the planet was still recorded as
the speediest machine in the November 2015 list. The PC beat out second place
finisher Titan by almost a 2 to1 edge. Titan, which is housed at the U.S.
Division of Energy's Oak Ridge National Laboratory, accomplished 17.59
petaflops, while Tianhe2 accomplished 33.86 petaflops. Tianhe2' s execution
gave back the title of the world's quickest supercomputer to China after
TianheI's début in November 2010. The Institute of Electrical and Electronics Engineers
said Tianhe2's win "symbolizes China's determined responsibility to the
supercomputing arms race". In June 2013, China housed 66 of the main 500
supercomputers, second just to the United States' 252 systems. The Chinese
aggregate expanded to 168 of the main 500 frameworks by June 2016, surpassing
the United States which tumbled to 165 of the main 500 supercomputers. Graph500
is a substitute rundown of top supercomputers in view of a benchmark testing
examination of graphs. In their benchmark, the framework tried at 2,061 Giga TEPS
(crossed edges every second). The top framework, IBM Sequoia, tried at 15,363 Giga
TEPS. It likewise has ahead of everyone else in the HPCG benchmark test
proposed by Jack Dongarra, with 0.580 HPCG PFLOPS in June 2014. Tianhe2 has
been housed at National University of Defense Technology.
Specification
As indicated by NUDT, Tianhe2 would have been
utilized for reproduction, examination, and Government security applications. With 16,000 PC hubs, each
including two Intel Ivy Bridge Xeon processors and three Xeon Phi coprocessor
chips, it spoke to the world's biggest establishment of Ivy Bridge and Xeon Phi
chips, numbering a sum of 3,120,000 cores. Each of the 16,000 hubs had 88
gigabytes of memory (64 utilized by the Ivy Bridge processors, and 8 gigabytes
for each of the Xeon Phi processors). The aggregate CPU in addition to
coprocessor memory was 1,375 TiB (roughly 1.34 PiB). The framework has a 12.4
PiB H2FS document framework comprising of IO sending hubs giving a 1 TiB/s
burst rate sponsored by a Luster document framework with 100 GiB/s maintained
through put. Amid the testing stage, Tianhe2 was laid out in a non optimal
limited space. At the point when gathered at its last area, the framework will
have had a hypothetical pinnacle execution of 54.9 pet flops. At pinnacle
control utilization, the frame work itself would have drawn 17.6 megawatts of
force. Counting outside cooling, the framework drew a total of 24 megawatts. 
The finished PC complex would have involved 720 square meters of space. The
frontend framework comprised of 4096 Galaxy FT1500 CPUs, a SPARC subordinate
outlined and worked by NUDT. Each FT1500
has 16 centers and a 1.8 GHz clock recurrence. The chip has an execution of 144
gigaflops and keeps running on 65 watts. The interconnect, called the TH
Express2, composed by NUDT, used a fat tree topology with 13 switches each of
576 ports. Tianhe2 kept running on Kylin
Linux, a rendition of the working framework created by NUDT. Asset
administration is based on Slurm Workload Manager.
0 comments:
Post a Comment