t**t
发帖数: 27760 | 1 MANNHEIM, Germany; BERKELEY, Calif.; and KNOXVILLE, Tenn.--Tianhe-2, a super
computer developed by China’s National University of Defense Technology, is
the world’s new No. 1 system with a performance of 33.86 petaflop/s on the
Linpack benchmark, according to the 41stedition of the twice-yearlyTOP500 l
ist of the world’s most powerful supercomputers. The list was announced Jun
e 17 during the opening session of the 2013 International Supercomputing Con
ference in Leipzig, Germany.
Tianhe-2, or Milky Way-2, will be deployed at the National Supercomputer Cen
ter in Guangzho, China, by the end of the year. The surprise appearance of T
ianhe-2, two years ahead of the expected deployment, marks China’s first re
turn to the No. 1 position since November 2010, when Tianhe-1A was the top s
ystem. Tianhe-2 has 16,000 nodes, each with two Intel Xeon IvyBridge process
ors and three Xeon Phi processors for a combined total of 3,120,000 computin
g cores.
Titan, a Cray XK7 system installed at the U.S. Department of Energy’s (DOE)
Oak Ridge National Laboratory and previously the No. 1 system, is now ranke
d No. 2. Titan achieved 17.59 petaflop/s on the Linpack benchmark using 261,
632 of its NVIDIA K20x accelerator cores. Titan is one of the most energy ef
ficient systems on the list, consuming a total of 8.21 MW and delivering 2,1
43 Mflops/W.
Sequoia, an IBM BlueGene/Q system installed at DOE’s Lawrence Livermore Nat
ional Laboratory, also dropped one position and is now the No. 3 system. Seq
uoia was first delivered in 2011 and has achieved 17.17 petaflop/s on the Li
npack benchmark using 1,572,864 cores. Sequoia is also one of the most energ
y efficient systems on the list, consuming a total of 7.84 MW and delivering
2,031.6 Mflops/W.
Fujitsu’s “K computer” installed at the RIKEN Advanced Institute for Comp
utational Science (AICS) in Kobe, Japan, is now the No. 4 system with a perf
ormance of 10.51 Pflop/s on the Linpack benchmark using 705,024 SPARC64 proc
essing cores
A second BlueGene/Q system, Mira, installed at Argonne National Laboratory i
s at No. 5 with 8.59 petaflop/s on the Linpack benchmark using 786,432 cores
.
Rounding out the Top 10 are the upgraded Stampede at the Texas Advanced Comp
uting Center of the University of Texas, Austin; JUQUEEN at the Forschungsze
ntrum Juelich in Germany (and the most powerful system in Europe); an IBM Bl
ueGene/Q system at Lawrence Livermore National Laboratory; SuperMUC, an IBM
iDataplex system installed at Leibniz Rechenzentrum in Germany; and Tianhe-1
A at the National Supercomputing Center in Tianjin, China.
According to TOP500 Editor Jack Dongarra, who toured the Tianhe-2 developmen
t facility in May, the system is noteworthy in a number of aspects. “Most o
f the features of the system were developed in China, and they are only usin
g Intel for the main compute part. That is, the interconnect, operating syst
em, front-end processors and software are mainly Chinese,” Dongarra said.
Other highlights from the June 2013 TOP500 List, which can be found at www.t
op500.org, include:
?There are 26 systems with performance greater than a petaflop/s, up from 23
six months ago.
?The new No. 1 system, Tianhe-2, and the No. 6 system, Stampede, are using I
ntel Xeon Phi processors to speed up their computational rate. The No. 2 sys
tem, Titan, and the No. 10 system, Tianhe-1A, are using NVIDIA GPUs to accel
erate computation.
? A total of 54 systems on the list are using accelerator/co-processor techn
ology, down from 62 in November 2012. Thirty-nine of these use NVIDIA chips,
three use ATI Radeon, and eleven systems use Intel MIC technology (Xeon Phi
).
?The number of systems installed in China has now stabilized at 66, with 72
and 68 on the last two lists. As a nation, China now holds the No. 2 positio
n as a user of HPC, ahead of Japan, UK, France, and Germany. Due to Tianhe-2
, China has also taken the No. 2 position in the performance share, ahead of
Japan.
?Intel continues to provide the processors for the largest share (80.4 perce
nt) of TOP500 systems.
?Eighty-eight percent of the systems use processors with six or more cores,
and 67 percent with eight or more cores.
?IBM’s BlueGene/Q is still the most popular system in the TOP10 with four e
ntries, including No. 3, 5, 7, and 8.
?The entry level on the list moved up to the 96.6 teraflop/s mark on the Lin
pack benchmark, compared to 76.5 Tflop/s six months ago.
?The last system on the newest list was listed at position 322 in the previo
us TOP500 just six months ago.
?Total combined performance of all 500 systems has grown to 223 Pflop/s, com
pared to 162 Pflop/s six months ago and 123 Pflop/s one year ago.
Geographical observations
?The U.S. is clearly the leading consumer of HPC systems, with 252 of the 50
0 systems (250 last time). The European share (112 systems – 105 last time)
is still lower than the Asian share (119 systems – 123 last time).
?Dominant countries in Asia are China with 66 systems (down from 72) and Jap
an with 30 systems (down from 32).
?In Europe, UK, France, and Germany, are almost equal with 29, 23, and 19 re
spectively.
About the TOP500 ListThe first version of what became today’s TOP500 list s
tarted as an exercise for a small conference in Germany in June 1993. Out of
curiosity, the authors decided to revisit the list in November 1993 to see
how things had changed. About that time they realized they might be on to so
mething and decided to continue compiling the list, which is now a much-anti
cipated, much-watched and much-debated twice-yearly event.
The TOP500 list is compiled by Hans Meuer of the University of Mannheim, Ger
many; Erich Strohmaier and Horst Simon of Lawrence Berkeley National Laborat
ory; and Jack Dongarra of the University of Tennessee, Knoxville. | n******7
发帖数: 12463 | 2 这回不是gpu堆出来的了，还double了第二名的成绩，很牛啊
那个志强三核是什么东西？
super
is
the
l
Jun
Con
Cen
T
re
【在 t**t 的大作中提到】
: MANNHEIM, Germany; BERKELEY, Calif.; and KNOXVILLE, Tenn.--Tianhe-2, a super
: computer developed by China’s National University of Defense Technology, is
: the world’s new No. 1 system with a performance of 33.86 petaflop/s on the
: Linpack benchmark, according to the 41stedition of the twice-yearlyTOP500 l
: ist of the world’s most powerful supercomputers. The list was announced Jun
: e 17 during the opening session of the 2013 International Supercomputing Con
: ference in Leipzig, Germany.
: Tianhe-2, or Milky Way-2, will be deployed at the National Supercomputer Cen
: ter in Guangzho, China, by the end of the year. The surprise appearance of T
: ianhe-2, two years ahead of the expected deployment, marks China’s first re
| f****a
发帖数: 4708 | 3 Xeon Phi processors
intel的gpu wannabe，市场不接受后，换个马甲主攻HPC。 | n******7
发帖数: 12463 | 4 原来也是gpu，难怪
【在 f****a 的大作中提到】
: Xeon Phi processors
: intel的gpu wannabe，市场不接受后，换个马甲主攻HPC。
| t**t
发帖数: 27760 | 5 这个是堆GPU的，不过是Intel自己的而已。
【在 n******7 的大作中提到】
: 这回不是gpu堆出来的了，还double了第二名的成绩，很牛啊
: 那个志强三核是什么东西？
:
: super
: is
: the
: l
: Jun
: Con
: Cen
| t**t
发帖数: 27760 | 6 Intel还是牛。
【在 f****a 的大作中提到】
: Xeon Phi processors
: intel的gpu wannabe，市场不接受后，换个马甲主攻HPC。
| a***e
发帖数: 27968 | 7 XEON phi is X86, based on good old pentium, with 512SIMD
the latest could do over 1.0TFLOPS DP sustain rate.
even the gefore titan could only do 1.3TFLOPS DP peak.
quite impressive stuff.
【在 t**t 的大作中提到】
: 这个是堆GPU的，不过是Intel自己的而已。
| l***j
发帖数: 300 | | a***e
发帖数: 27968 | 9 丫其实裸算性能很强劲，5年前就是单片现在titan的双精度性能
估计一些专用的材质，渲染运算不行，不适合当GPU
话说丫512SIMD和龙芯一个路子，1.2GHz, 61核就能到1.2T的性能
每个核一周期16个双精度运算，512/64=8,估计是FMA算俩了
【在 l***j 的大作中提到】
: Xeon Phi is Larrabee
| G*****h
发帖数: 33134 | 10 这种除了矩阵运算牛鼻还能干啥子用
【在 a***e 的大作中提到】
: 丫其实裸算性能很强劲，5年前就是单片现在titan的双精度性能
: 估计一些专用的材质，渲染运算不行，不适合当GPU
: 话说丫512SIMD和龙芯一个路子，1.2GHz, 61核就能到1.2T的性能
: 每个核一周期16个双精度运算，512/64=8,估计是FMA算俩了
| | | t**t
发帖数: 27760 | 11 比如算算你的买买提密码
【在 G*****h 的大作中提到】
: 这种除了矩阵运算牛鼻还能干啥子用
| G*****h
发帖数: 33134 | 12 算密码不一定牛吧
只适合穷举法
不知道搞有限元是不是很好
【在 t**t 的大作中提到】
: 比如算算你的买买提密码
| y**b
发帖数: 10166 | 13 牛，看了一下说明感觉比gpu好用。最重要的是原来x86的代码不管用改写吧？
数据带宽也高得多。
60cores，240threads，还真是有点强大，不知道主要弱点在哪里？openmp
估计也会加点新东西了。
【在 a***e 的大作中提到】
: XEON phi is X86, based on good old pentium, with 512SIMD
: the latest could do over 1.0TFLOPS DP sustain rate.
: even the gefore titan could only do 1.3TFLOPS DP peak.
: quite impressive stuff.
| f****a
发帖数: 4708 | 14 原来的x86代码如果用了512b Larrabee指令的话，倒是不用改写。
【在 y**b 的大作中提到】
: 牛，看了一下说明感觉比gpu好用。最重要的是原来x86的代码不管用改写吧？
: 数据带宽也高得多。
: 60cores，240threads，还真是有点强大，不知道主要弱点在哪里？openmp
: 估计也会加点新东西了。
| l***j
发帖数: 300 | 15 if your code is already multithreaded, you will get performance boost
automatically without rewrite your code. if you convert your work load to
use Larrabee Instructions, you will get instruction level parallelism.
【在 f****a 的大作中提到】
: 原来的x86代码如果用了512b Larrabee指令的话，倒是不用改写。
|
|
