<s>
A	O
supercomputer	B-Architecture
is	O
a	O
computer	O
with	O
a	O
high	O
level	O
of	O
performance	O
as	O
compared	O
to	O
a	O
general-purpose	O
computer	O
.	O
</s>
<s>
The	O
performance	O
of	O
a	O
supercomputer	B-Architecture
is	O
commonly	O
measured	O
in	O
floating-point	B-Algorithm
operations	O
per	O
second	O
(	O
FLOPS	O
)	O
instead	O
of	O
million	O
instructions	O
per	O
second	O
(	O
MIPS	O
)	O
.	O
</s>
<s>
Since	O
2017	O
,	O
there	O
have	O
existed	O
supercomputers	B-Architecture
which	O
can	O
perform	O
over	O
1017FLOPS	O
(	O
a	O
hundred	O
quadrillion	O
FLOPS	O
,	O
100petaFLOPS	O
or	O
100PFLOPS	O
)	O
.	O
</s>
<s>
Since	O
November	O
2017	O
,	O
all	O
of	O
the	O
world	B-Operating_System
's	I-Operating_System
fastest	I-Operating_System
500	I-Operating_System
supercomputers	I-Operating_System
run	O
on	O
Linux-based	O
operating	B-General_Concept
systems	I-General_Concept
.	O
</s>
<s>
Additional	O
research	O
is	O
being	O
conducted	O
in	O
the	O
United	O
States	O
,	O
the	O
European	O
Union	O
,	O
Taiwan	O
,	O
Japan	O
,	O
and	O
China	O
to	O
build	O
faster	O
,	O
more	O
powerful	O
and	O
technologically	O
superior	O
exascale	B-General_Concept
supercomputers	I-General_Concept
.	O
</s>
<s>
Supercomputers	B-Architecture
play	O
an	O
important	O
role	O
in	O
the	O
field	O
of	O
computational	O
science	O
,	O
and	O
are	O
used	O
for	O
a	O
wide	O
range	O
of	O
computationally	O
intensive	O
tasks	O
in	O
various	O
fields	O
,	O
including	O
quantum	O
mechanics	O
,	O
weather	O
forecasting	O
,	O
climate	O
research	O
,	O
oil	O
and	O
gas	O
exploration	O
,	O
molecular	O
modeling	O
(	O
computing	O
the	O
structures	O
and	O
properties	O
of	O
chemical	O
compounds	O
,	O
biological	O
macromolecules	O
,	O
polymers	O
,	O
and	O
crystals	O
)	O
,	O
and	O
physical	O
simulations	O
(	O
such	O
as	O
simulations	O
of	O
the	O
early	O
moments	O
of	O
the	O
universe	O
,	O
airplane	O
and	O
spacecraft	O
aerodynamics	O
,	O
the	O
detonation	O
of	O
nuclear	O
weapons	O
,	O
and	O
nuclear	O
fusion	O
)	O
.	O
</s>
<s>
Supercomputers	B-Architecture
were	O
introduced	O
in	O
the	O
1960s	O
,	O
and	O
for	O
several	O
decades	O
the	O
fastest	O
were	O
made	O
by	O
Seymour	O
Cray	O
at	O
Control	O
Data	O
Corporation	O
(	O
CDC	O
)	O
,	O
Cray	O
Research	O
and	O
subsequent	O
companies	O
bearing	O
his	O
name	O
or	O
monogram	O
.	O
</s>
<s>
Through	O
the	O
decade	O
,	O
increasing	O
amounts	O
of	O
parallelism	B-Operating_System
were	O
added	O
,	O
with	O
one	O
to	O
four	O
processors	B-General_Concept
being	O
typical	O
.	O
</s>
<s>
In	O
the	O
1970s	O
,	O
vector	B-Operating_System
processors	I-Operating_System
operating	O
on	O
large	O
arrays	O
of	O
data	O
came	O
to	O
dominate	O
.	O
</s>
<s>
A	O
notable	O
example	O
is	O
the	O
highly	O
successful	O
Cray-1	B-Device
of	O
1976	O
.	O
</s>
<s>
Vector	B-Operating_System
computers	I-Operating_System
remained	O
the	O
dominant	O
design	O
into	O
the	O
1990s	O
.	O
</s>
<s>
From	O
then	O
until	O
today	O
,	O
massively	B-Operating_System
parallel	I-Operating_System
supercomputers	B-Architecture
with	O
tens	O
of	O
thousands	O
of	O
off-the-shelf	O
processors	B-General_Concept
became	O
the	O
norm	O
.	O
</s>
<s>
The	O
US	O
has	O
long	O
been	O
the	O
leader	O
in	O
the	O
supercomputer	B-Architecture
field	O
,	O
first	O
through	O
Cray	O
's	O
almost	O
uninterrupted	O
dominance	O
of	O
the	O
field	O
,	O
and	O
later	O
through	O
a	O
variety	O
of	O
technology	O
companies	O
.	O
</s>
<s>
As	O
of	O
May	O
2022	O
,	O
the	O
fastest	B-Operating_System
supercomputer	I-Operating_System
on	O
the	O
TOP500	B-Operating_System
supercomputer	B-Architecture
list	O
is	O
Frontier	B-Device
,	O
in	O
the	O
US	O
,	O
with	O
a	O
LINPACK	B-Device
benchmark	I-Device
score	O
of	O
1.102	O
ExaFlop/s	O
,	O
followed	O
by	O
Fugaku	B-Device
.	O
</s>
<s>
In	O
June	O
2018	O
,	O
all	O
combined	O
supercomputers	B-Architecture
on	O
the	O
TOP500	B-Operating_System
list	I-Operating_System
broke	O
the	O
1exaFLOPS	O
mark	O
.	O
</s>
<s>
In	O
1960	O
,	O
UNIVAC	O
built	O
the	O
Livermore	B-Device
Atomic	I-Device
Research	I-Device
Computer	I-Device
(	O
LARC	O
)	O
,	O
today	O
considered	O
among	O
the	O
first	O
supercomputers	B-Architecture
,	O
for	O
the	O
US	O
Navy	O
Research	O
and	O
Development	O
Center	O
.	O
</s>
<s>
It	O
still	O
used	O
high-speed	O
drum	B-General_Concept
memory	I-General_Concept
,	O
rather	O
than	O
the	O
newly	O
emerging	O
disk	B-Device
drive	I-Device
technology	O
.	O
</s>
<s>
Also	O
,	O
among	O
the	O
first	O
supercomputers	B-Architecture
was	O
the	O
IBM	B-Device
7030	I-Device
Stretch	I-Device
.	O
</s>
<s>
The	O
IBM	B-Device
7030	I-Device
was	O
built	O
by	O
IBM	O
for	O
the	O
Los	O
Alamos	O
National	O
Laboratory	O
,	O
which	O
in	O
1955	O
had	O
requested	O
a	O
computer	O
100	O
times	O
faster	O
than	O
any	O
existing	O
computer	O
.	O
</s>
<s>
The	O
IBM	B-Device
7030	I-Device
used	O
transistors	B-Device
,	O
magnetic	O
core	O
memory	O
,	O
pipelined	B-General_Concept
instructions	O
,	O
prefetched	O
data	O
through	O
a	O
memory	O
controller	O
and	O
included	O
pioneering	O
random	O
access	O
disk	B-Device
drives	I-Device
.	O
</s>
<s>
The	O
IBM	B-Device
7030	I-Device
was	O
completed	O
in	O
1961	O
and	O
despite	O
not	O
meeting	O
the	O
challenge	O
of	O
a	O
hundredfold	O
increase	O
in	O
performance	O
,	O
it	O
was	O
purchased	O
by	O
the	O
Los	O
Alamos	O
National	O
Laboratory	O
.	O
</s>
<s>
Customers	O
in	O
England	O
and	O
France	O
also	O
bought	O
the	O
computer	O
,	O
and	O
it	O
became	O
the	O
basis	O
for	O
the	O
IBM	B-Device
7950	I-Device
Harvest	I-Device
,	O
a	O
supercomputer	B-Architecture
built	O
for	O
cryptanalysis	O
.	O
</s>
<s>
The	O
third	O
pioneering	O
supercomputer	B-Architecture
project	O
in	O
the	O
early	O
1960s	O
was	O
the	O
Atlas	B-Device
at	O
the	O
University	O
of	O
Manchester	O
,	O
built	O
by	O
a	O
team	O
led	O
by	O
Tom	O
Kilburn	O
.	O
</s>
<s>
He	O
designed	O
the	O
Atlas	B-Device
to	O
have	O
memory	O
space	O
for	O
up	O
to	O
a	O
million	O
words	O
of	O
48	O
bits	O
,	O
but	O
because	O
magnetic	O
storage	O
with	O
such	O
a	O
capacity	O
was	O
unaffordable	O
,	O
the	O
actual	O
core	O
memory	O
of	O
the	O
Atlas	B-Device
was	O
only	O
16,000	O
words	O
,	O
with	O
a	O
drum	O
providing	O
memory	O
for	O
a	O
further	O
96,000	O
words	O
.	O
</s>
<s>
The	O
Atlas	B-Device
operating	B-General_Concept
system	I-General_Concept
swapped	O
data	O
in	O
the	O
form	O
of	O
pages	O
between	O
the	O
magnetic	O
core	O
and	O
the	O
drum	O
.	O
</s>
<s>
The	O
Atlas	B-Device
operating	B-General_Concept
system	I-General_Concept
also	O
introduced	O
time-sharing	B-General_Concept
to	O
supercomputing	B-Architecture
,	O
so	O
that	O
more	O
than	O
one	O
program	O
could	O
be	O
executed	O
on	O
the	O
supercomputer	B-Architecture
at	O
any	O
one	O
time	O
.	O
</s>
<s>
Atlas	B-Device
was	O
a	O
joint	O
venture	O
between	O
Ferranti	O
and	O
the	O
Manchester	O
University	O
and	O
was	O
designed	O
to	O
operate	O
at	O
processing	O
speeds	O
approaching	O
onemicrosecond	O
per	O
instruction	O
,	O
about	O
onemillion	O
instructions	O
per	O
second	O
.	O
</s>
<s>
The	O
CDC	B-Device
6600	I-Device
,	O
designed	O
by	O
Seymour	O
Cray	O
,	O
was	O
finished	O
in	O
1964	O
and	O
marked	O
the	O
transition	O
from	O
germanium	O
to	O
silicon	O
transistors	B-Device
.	O
</s>
<s>
Silicon	O
transistors	B-Device
could	O
run	O
more	O
quickly	O
and	O
the	O
overheating	O
problem	O
was	O
solved	O
by	O
introducing	O
refrigeration	O
to	O
the	O
supercomputer	B-Architecture
design	O
.	O
</s>
<s>
Thus	O
,	O
the	O
CDC6600	O
became	O
the	O
fastest	B-Operating_System
computer	I-Operating_System
in	O
the	O
world	O
.	O
</s>
<s>
Given	O
that	O
the	O
6600	O
outperformed	O
all	O
the	O
other	O
contemporary	O
computers	O
by	O
about	O
10	O
times	O
,	O
it	O
was	O
dubbed	O
a	O
supercomputer	B-Architecture
and	O
defined	O
the	O
supercomputing	B-Architecture
market	O
,	O
when	O
one	O
hundred	O
computers	O
were	O
sold	O
at	O
$8	O
million	O
each	O
.	O
</s>
<s>
Four	O
years	O
after	O
leaving	O
CDC	O
,	O
Cray	O
delivered	O
the	O
80MHz	O
Cray-1	B-Device
in	O
1976	O
,	O
which	O
became	O
one	O
of	O
the	O
most	O
successful	O
supercomputers	B-Architecture
in	O
history	O
.	O
</s>
<s>
The	O
Cray-2	B-Device
was	O
released	O
in	O
1985	O
.	O
</s>
<s>
It	O
had	O
eight	O
central	B-General_Concept
processing	I-General_Concept
units	I-General_Concept
(	O
CPUs	O
)	O
,	O
liquid	B-General_Concept
cooling	I-General_Concept
and	O
the	O
electronics	O
coolant	O
liquid	O
Fluorinert	O
was	O
pumped	O
through	O
the	O
supercomputer	B-Operating_System
architecture	I-Operating_System
.	O
</s>
<s>
It	O
reached	O
1.9gigaFLOPS	O
,	O
making	O
it	O
the	O
first	O
supercomputer	B-Architecture
to	O
break	O
the	O
gigaflop	O
barrier	O
.	O
</s>
<s>
The	O
only	O
computer	O
to	O
seriously	O
challenge	O
the	O
Cray-1	B-Device
'	O
s	O
performance	O
in	O
the	O
1970s	O
was	O
the	O
ILLIAC	B-Device
IV	I-Device
.	O
</s>
<s>
This	O
machine	O
was	O
the	O
first	O
realized	O
example	O
of	O
a	O
true	O
massively	B-Operating_System
parallel	I-Operating_System
computer	I-Operating_System
,	O
in	O
which	O
many	O
processors	B-General_Concept
worked	O
together	O
to	O
solve	O
different	O
parts	O
of	O
a	O
single	O
larger	O
problem	O
.	O
</s>
<s>
In	O
contrast	O
with	O
the	O
vector	O
systems	O
,	O
which	O
were	O
designed	O
to	O
run	O
a	O
single	O
stream	O
of	O
data	O
as	O
quickly	O
as	O
possible	O
,	O
in	O
this	O
concept	O
,	O
the	O
computer	O
instead	O
feeds	O
separate	O
parts	O
of	O
the	O
data	O
to	O
entirely	O
different	O
processors	B-General_Concept
and	O
then	O
recombines	O
the	O
results	O
.	O
</s>
<s>
The	O
ILLIAC	O
's	O
design	O
was	O
finalized	O
in	O
1966	O
with	O
256	O
processors	B-General_Concept
and	O
offer	O
speed	O
up	O
to	O
1GFLOPS	O
,	O
compared	O
to	O
the	O
1970s	O
Cray-1	B-Device
'	O
s	O
peak	O
of	O
250MFLOPS	O
.	O
</s>
<s>
However	O
,	O
development	O
problems	O
led	O
to	O
only	O
64	O
processors	B-General_Concept
being	O
built	O
,	O
and	O
the	O
system	O
could	O
never	O
operate	O
more	O
quickly	O
than	O
about	O
200MFLOPS	O
while	O
being	O
much	O
larger	O
and	O
more	O
complex	O
than	O
the	O
Cray	O
.	O
</s>
<s>
But	O
the	O
partial	O
success	O
of	O
the	O
ILLIAC	B-Device
IV	I-Device
was	O
widely	O
seen	O
as	O
pointing	O
the	O
way	O
to	O
the	O
future	O
of	O
supercomputing	B-Architecture
.	O
</s>
<s>
But	O
by	O
the	O
early	O
1980s	O
,	O
several	O
teams	O
were	O
working	O
on	O
parallel	O
designs	O
with	O
thousands	O
of	O
processors	B-General_Concept
,	O
notably	O
the	B-Device
Connection	I-Device
Machine	I-Device
(	O
CM	O
)	O
that	O
developed	O
from	O
research	O
at	O
MIT	O
.	O
</s>
<s>
The	O
CM-1	O
used	O
as	O
many	O
as	O
65,536	O
simplified	O
custom	O
microprocessors	B-Architecture
connected	O
together	O
in	O
a	O
network	B-Architecture
to	O
share	O
data	O
.	O
</s>
<s>
Several	O
updated	O
versions	O
followed	O
;	O
the	O
CM-5	B-Device
supercomputer	B-Architecture
is	O
a	O
massively	B-Operating_System
parallel	I-Operating_System
processing	I-Operating_System
computer	O
capable	O
of	O
many	O
billions	O
of	O
arithmetic	O
operations	O
per	O
second	O
.	O
</s>
<s>
In	O
1982	O
,	O
Osaka	O
University	O
's	O
LINKS-1	B-Device
Computer	I-Device
Graphics	I-Device
System	I-Device
used	O
a	O
massively	B-Operating_System
parallel	I-Operating_System
processing	I-Operating_System
architecture	O
,	O
with	O
514	O
microprocessors	B-Architecture
,	O
including	O
257	O
Zilog	B-Device
Z8001	I-Device
control	B-General_Concept
processors	I-General_Concept
and	O
257	O
iAPX	B-Device
86/20	I-Device
floating-point	B-General_Concept
processors	I-General_Concept
.	O
</s>
<s>
Fujitsu	O
's	O
VPP500	O
from	O
1992	O
is	O
unusual	O
since	O
,	O
to	O
achieve	O
higher	O
speeds	O
,	O
its	O
processors	B-General_Concept
used	O
GaAs	O
,	O
a	O
material	O
normally	O
reserved	O
for	O
microwave	O
applications	O
due	O
to	O
its	O
toxicity	O
.	O
</s>
<s>
Fujitsu	O
's	O
Numerical	B-Device
Wind	I-Device
Tunnel	I-Device
supercomputer	B-Architecture
used	O
166	O
vector	B-Operating_System
processors	I-Operating_System
to	O
gain	O
the	O
top	O
spot	O
in	O
1994	O
with	O
a	O
peak	O
speed	O
of	O
1.7gigaFLOPS	O
(	O
GFLOPS	O
)	O
per	O
processor	O
.	O
</s>
<s>
The	O
Hitachi	O
SR2201	O
obtained	O
a	O
peak	O
performance	O
of	O
600GFLOPS	O
in	O
1996	O
by	O
using	O
2048	O
processors	B-General_Concept
connected	O
via	O
a	O
fast	O
three-dimensional	O
crossbar	O
network	B-Architecture
.	O
</s>
<s>
The	O
Intel	B-Device
Paragon	I-Device
could	O
have	O
1000	O
to	O
4000	O
Intel	B-General_Concept
i860	I-General_Concept
processors	B-General_Concept
in	O
various	O
configurations	O
and	O
was	O
ranked	O
the	O
fastest	O
in	O
the	O
world	O
in	O
1993	O
.	O
</s>
<s>
The	O
Paragon	O
was	O
a	O
MIMD	B-Operating_System
machine	O
which	O
connected	O
processors	B-General_Concept
via	O
a	O
high	O
speed	O
two-dimensional	O
mesh	O
,	O
allowing	O
processes	O
to	O
execute	O
on	O
separate	O
nodes	B-Operating_System
,	O
communicating	O
via	O
the	O
Message	B-Application
Passing	I-Application
Interface	I-Application
.	O
</s>
<s>
Similar	O
designs	O
using	O
custom	O
hardware	O
were	O
made	O
by	O
many	O
companies	O
,	O
including	O
the	O
Evans	B-Device
&	I-Device
Sutherland	I-Device
ES-1	I-Device
,	O
MasPar	B-Device
,	O
nCUBE	B-Application
,	O
Intel	B-Device
iPSC	I-Device
and	O
the	O
Goodyear	B-Device
MPP	I-Device
.	O
</s>
<s>
But	O
by	O
the	O
mid-1990s	O
,	O
general-purpose	O
CPU	O
performance	O
had	O
improved	O
so	O
much	O
in	O
that	O
a	O
supercomputer	B-Architecture
could	O
be	O
built	O
using	O
them	O
as	O
the	O
individual	O
processing	O
units	O
,	O
instead	O
of	O
using	O
custom	O
chips	O
.	O
</s>
<s>
By	O
the	O
turn	O
of	O
the	O
21st	O
century	O
,	O
designs	O
featuring	O
tens	O
of	O
thousands	O
of	O
commodity	O
CPUs	O
were	O
the	O
norm	O
,	O
with	O
later	O
machines	O
adding	O
graphic	B-Architecture
units	I-Architecture
to	O
the	O
mix	O
.	O
</s>
<s>
Systems	O
with	O
a	O
massive	O
number	O
of	O
processors	B-General_Concept
generally	O
take	O
one	O
of	O
two	O
paths	O
.	O
</s>
<s>
In	O
the	O
grid	B-Architecture
computing	I-Architecture
approach	O
,	O
the	O
processing	O
power	O
of	O
many	O
computers	O
,	O
organized	O
as	O
distributed	O
,	O
diverse	O
administrative	O
domains	O
,	O
is	O
opportunistically	O
used	O
whenever	O
a	O
computer	O
is	O
available	O
.	O
</s>
<s>
In	O
another	O
approach	O
,	O
many	O
processors	B-General_Concept
are	O
used	O
in	O
proximity	O
to	O
each	O
other	O
,	O
e.g.	O
</s>
<s>
in	O
a	O
computer	B-Architecture
cluster	I-Architecture
.	O
</s>
<s>
In	O
such	O
a	O
centralized	O
massively	B-Operating_System
parallel	I-Operating_System
system	O
the	O
speed	O
and	O
flexibility	O
of	O
the	O
becomes	O
very	O
important	O
and	O
modern	O
supercomputers	B-Architecture
have	O
used	O
various	O
approaches	O
ranging	O
from	O
enhanced	O
Infiniband	B-Architecture
systems	O
to	O
three-dimensional	O
torus	B-Operating_System
interconnects	I-Operating_System
.	O
</s>
<s>
The	O
use	O
of	O
multi-core	B-Architecture
processors	I-Architecture
combined	O
with	O
centralization	O
is	O
an	O
emerging	O
direction	O
,	O
e.g.	O
</s>
<s>
as	O
in	O
the	O
Cyclops64	B-Device
system	O
.	O
</s>
<s>
As	O
the	O
price	O
,	O
performance	O
and	O
energy	O
efficiency	O
of	O
general-purpose	B-Architecture
graphics	I-Architecture
processing	I-Architecture
units	I-Architecture
(	O
GPGPUs	B-Architecture
)	O
have	O
improved	O
,	O
a	O
number	O
of	O
petaFLOPS	O
supercomputers	B-Architecture
such	O
as	O
Tianhe-I	B-Device
and	O
Nebulae	B-Device
have	O
started	O
to	O
rely	O
on	O
them	O
.	O
</s>
<s>
However	O
,	O
other	O
systems	O
such	O
as	O
the	O
K	B-Device
computer	I-Device
continue	O
to	O
use	O
conventional	O
processors	B-General_Concept
such	O
as	O
SPARC-based	O
designs	O
and	O
the	O
overall	O
applicability	O
of	O
GPGPUs	B-Architecture
in	O
general-purpose	O
high-performance	B-Architecture
computing	I-Architecture
applications	O
has	O
been	O
the	O
subject	O
of	O
debate	O
,	O
in	O
that	O
while	O
a	O
GPGPU	B-Architecture
may	O
be	O
tuned	O
to	O
score	O
well	O
on	O
specific	O
benchmarks	O
,	O
its	O
overall	O
applicability	O
to	O
everyday	O
algorithms	O
may	O
be	O
limited	O
unless	O
significant	O
effort	O
is	O
spent	O
to	O
tune	O
the	O
application	O
to	O
it	O
.	O
</s>
<s>
However	O
,	O
GPUs	O
are	O
gaining	O
ground	O
,	O
and	O
in	O
2012	O
the	O
Jaguar	B-Device
supercomputer	I-Device
was	O
transformed	O
into	O
Titan	B-Device
by	O
retrofitting	O
CPUs	O
with	O
GPUs	O
.	O
</s>
<s>
High-performance	B-Architecture
computers	I-Architecture
have	O
an	O
expected	O
life	O
cycle	O
of	O
about	O
three	O
years	O
before	O
requiring	O
an	O
upgrade	O
.	O
</s>
<s>
The	O
Gyoukou	O
supercomputer	B-Architecture
is	O
unique	O
in	O
that	O
it	O
uses	O
both	O
a	O
massively	B-Operating_System
parallel	I-Operating_System
design	O
and	O
liquid	B-Device
immersion	I-Device
cooling	I-Device
.	O
</s>
<s>
This	O
allows	O
the	O
use	O
of	O
specially	O
programmed	O
FPGA	B-Architecture
chips	O
or	O
even	O
custom	O
ASICs	O
,	O
allowing	O
better	O
price/performance	O
ratios	O
by	O
sacrificing	O
generality	O
.	O
</s>
<s>
Examples	O
of	O
special-purpose	O
supercomputers	B-Architecture
include	O
Belle	B-General_Concept
,	O
Deep	B-General_Concept
Blue	I-General_Concept
,	O
and	O
Hydra	B-Architecture
for	O
playing	O
chess	B-Application
,	O
Gravity	B-Device
Pipe	I-Device
for	O
astrophysics	O
,	O
MDGRAPE-3	B-Algorithm
for	O
protein	O
structure	O
prediction	O
and	O
molecular	O
dynamics	O
,	O
and	O
Deep	B-Algorithm
Crack	I-Algorithm
for	O
breaking	O
the	O
DES	B-Algorithm
cipher	B-Application
.	O
</s>
<s>
Throughout	O
the	O
decades	O
,	O
the	O
management	O
of	O
heat	O
density	O
has	O
remained	O
a	O
key	O
issue	O
for	O
most	O
centralized	O
supercomputers	B-Architecture
.	O
</s>
<s>
A	O
typical	O
supercomputer	B-Architecture
consumes	O
large	O
amounts	O
of	O
electrical	O
power	O
,	O
almost	O
all	O
of	O
which	O
is	O
converted	O
into	O
heat	O
,	O
requiring	O
cooling	O
.	O
</s>
<s>
The	O
thermal	B-General_Concept
design	I-General_Concept
power	I-General_Concept
and	O
CPU	B-General_Concept
power	I-General_Concept
dissipation	I-General_Concept
issues	O
in	O
supercomputing	B-Architecture
surpass	O
those	O
of	O
traditional	O
computer	B-General_Concept
cooling	I-General_Concept
technologies	O
.	O
</s>
<s>
The	O
supercomputing	B-Architecture
awards	O
for	O
green	O
computing	O
reflect	O
this	O
issue	O
.	O
</s>
<s>
The	O
packing	O
of	O
thousands	O
of	O
processors	B-General_Concept
together	O
inevitably	O
generates	O
significant	O
amounts	O
of	O
heat	O
density	O
that	O
need	O
to	O
be	O
dealt	O
with	O
.	O
</s>
<s>
The	O
Cray-2	B-Device
was	O
liquid	B-General_Concept
cooled	I-General_Concept
,	O
and	O
used	O
a	O
Fluorinert	O
"	O
cooling	O
waterfall	O
"	O
which	O
was	O
forced	O
through	O
the	O
modules	O
under	O
pressure	O
.	O
</s>
<s>
However	O
,	O
the	O
submerged	O
liquid	B-General_Concept
cooling	I-General_Concept
approach	O
was	O
not	O
practical	O
for	O
the	O
multi-cabinet	O
systems	O
based	O
on	O
off-the-shelf	O
processors	B-General_Concept
,	O
and	O
in	O
System	B-Device
X	I-Device
a	O
special	O
cooling	O
system	O
that	O
combined	O
air	O
conditioning	O
with	O
liquid	B-General_Concept
cooling	I-General_Concept
was	O
developed	O
in	O
conjunction	O
with	O
the	O
Liebert	O
company	O
.	O
</s>
<s>
In	O
the	O
Blue	B-Operating_System
Gene	I-Operating_System
system	O
,	O
IBM	O
deliberately	O
used	O
low	O
power	O
processors	B-General_Concept
to	O
deal	O
with	O
heat	O
density	O
.	O
</s>
<s>
In	O
2008	O
,	O
Roadrunner	B-General_Concept
by	O
IBM	O
operated	O
at	O
3.76MFLOPS/W	O
.	O
</s>
<s>
In	O
November	O
2010	O
,	O
the	O
Blue	O
Gene/Q	O
reached	O
1,684	O
MFLOPS/W	O
and	O
in	O
June	O
2011	O
the	O
top	O
two	O
spots	O
on	O
the	O
Green	B-Application
500	I-Application
list	O
were	O
occupied	O
by	O
Blue	B-Operating_System
Gene	I-Operating_System
machines	O
in	O
New	O
York	O
(	O
one	O
achieving	O
2097MFLOPS/W	O
)	O
with	O
the	O
DEGIMA	B-Device
cluster	I-Device
in	O
Nagasaki	O
placing	O
third	O
with	O
1375MFLOPS/W	O
.	O
</s>
<s>
Because	O
copper	O
wires	O
can	O
transfer	O
energy	O
into	O
a	O
supercomputer	B-Architecture
with	O
much	O
higher	O
power	O
densities	O
than	O
forced	O
air	O
or	O
circulating	O
refrigerants	O
can	O
remove	O
waste	O
heat	O
,	O
the	O
ability	O
of	O
the	O
cooling	O
systems	O
to	O
remove	O
waste	O
heat	O
is	O
a	O
limiting	O
factor	O
.	O
</s>
<s>
,	O
many	O
existing	O
supercomputers	B-Architecture
have	O
more	O
infrastructure	O
capacity	O
than	O
the	O
actual	O
peak	O
demand	O
of	O
the	O
machine	O
designers	O
generally	O
conservatively	O
design	O
the	O
power	O
and	O
cooling	O
infrastructure	O
to	O
handle	O
more	O
than	O
the	O
theoretical	O
peak	O
electrical	O
power	O
consumed	O
by	O
the	O
supercomputer	B-Architecture
.	O
</s>
<s>
Designs	O
for	O
future	O
supercomputers	B-Architecture
are	O
power-limited	O
the	O
thermal	B-General_Concept
design	I-General_Concept
power	I-General_Concept
of	O
the	O
supercomputer	B-Architecture
as	O
a	O
whole	O
,	O
the	O
amount	O
that	O
the	O
power	O
and	O
cooling	O
infrastructure	O
can	O
handle	O
,	O
is	O
somewhat	O
more	O
than	O
the	O
expected	O
normal	O
power	O
consumption	O
,	O
but	O
less	O
than	O
the	O
theoretical	O
peak	O
power	O
consumption	O
of	O
the	O
electronic	O
hardware	O
.	O
</s>
<s>
Since	O
the	O
end	O
of	O
the	O
20th	O
century	O
,	O
supercomputer	B-Operating_System
operating	I-Operating_System
systems	I-Operating_System
have	O
undergone	O
major	O
transformations	O
,	O
based	O
on	O
the	O
changes	O
in	O
supercomputer	B-Operating_System
architecture	I-Operating_System
.	O
</s>
<s>
While	O
early	O
operating	B-General_Concept
systems	I-General_Concept
were	O
custom	O
tailored	O
to	O
each	O
supercomputer	B-Architecture
to	O
gain	O
speed	O
,	O
the	O
trend	O
has	O
been	O
to	O
move	O
away	O
from	O
in-house	O
operating	B-General_Concept
systems	I-General_Concept
to	O
the	O
adaptation	O
of	O
generic	O
software	O
such	O
as	O
Linux	B-Application
.	O
</s>
<s>
Since	O
modern	O
massively	B-Operating_System
parallel	I-Operating_System
supercomputers	B-Architecture
typically	O
separate	O
computations	O
from	O
other	O
services	O
by	O
using	O
multiple	O
types	O
of	O
nodes	B-Operating_System
,	O
they	O
usually	O
run	O
different	O
operating	B-General_Concept
systems	I-General_Concept
on	O
different	O
nodes	B-Operating_System
,	O
e.g.	O
</s>
<s>
using	O
a	O
small	O
and	O
efficient	O
lightweight	B-Device
kernel	I-Device
such	O
as	O
CNK	B-Operating_System
or	O
CNL	B-Application
on	O
compute	O
nodes	B-Operating_System
,	O
but	O
a	O
larger	O
system	O
such	O
as	O
a	O
Linux-derivative	O
on	O
server	O
and	O
I/O	B-General_Concept
nodes	B-Operating_System
.	O
</s>
<s>
While	O
in	O
a	O
traditional	O
multi-user	O
computer	O
system	O
job	B-General_Concept
scheduling	I-General_Concept
is	O
,	O
in	O
effect	O
,	O
a	O
tasking	O
problem	O
for	O
processing	O
and	O
peripheral	O
resources	O
,	O
in	O
a	O
massively	B-Operating_System
parallel	I-Operating_System
system	O
,	O
the	O
job	O
management	O
system	O
needs	O
to	O
manage	O
the	O
allocation	O
of	O
both	O
computational	O
and	O
communication	O
resources	O
,	O
as	O
well	O
as	O
gracefully	O
deal	O
with	O
inevitable	O
hardware	O
failures	O
when	O
tens	O
of	O
thousands	O
of	O
processors	B-General_Concept
are	O
present	O
.	O
</s>
<s>
Although	O
most	O
modern	O
supercomputers	B-Architecture
use	O
Linux-based	O
operating	B-General_Concept
systems	I-General_Concept
,	O
each	O
manufacturer	O
has	O
its	O
own	O
specific	O
Linux-derivative	O
,	O
and	O
no	O
industry	O
standard	O
exists	O
,	O
partly	O
due	O
to	O
the	O
fact	O
that	O
the	O
differences	O
in	O
hardware	O
architectures	O
require	O
changes	O
to	O
optimize	O
the	O
operating	B-General_Concept
system	I-General_Concept
to	O
each	O
hardware	O
design	O
.	O
</s>
<s>
The	O
parallel	B-Operating_System
architectures	I-Operating_System
of	O
supercomputers	B-Architecture
often	O
dictate	O
the	O
use	O
of	O
special	O
programming	O
techniques	O
to	O
exploit	O
their	O
speed	O
.	O
</s>
<s>
Software	O
tools	O
for	O
distributed	B-Architecture
processing	I-Architecture
include	O
standard	O
APIs	B-Application
such	O
as	O
MPI	B-Application
and	O
PVM	B-Operating_System
,	O
VTL	B-Device
,	O
and	O
open	B-Application
source	I-Application
software	I-Application
such	O
as	O
Beowulf	B-Operating_System
.	O
</s>
<s>
In	O
the	O
most	O
common	O
scenario	O
,	O
environments	O
such	O
as	O
PVM	B-Operating_System
and	O
MPI	B-Application
for	O
loosely	O
connected	O
clusters	O
and	O
OpenMP	B-Application
for	O
tightly	O
coordinated	O
shared	O
memory	O
machines	O
are	O
used	O
.	O
</s>
<s>
Significant	O
effort	O
is	O
required	O
to	O
optimize	O
an	O
algorithm	O
for	O
the	O
interconnect	O
characteristics	O
of	O
the	O
machine	O
it	O
will	O
be	O
run	O
on	O
;	O
the	O
aim	O
is	O
to	O
prevent	O
any	O
of	O
the	O
CPUs	O
from	O
wasting	O
time	O
waiting	O
on	O
data	O
from	O
other	O
nodes	B-Operating_System
.	O
</s>
<s>
GPGPUs	B-Architecture
have	O
hundreds	O
of	O
processor	O
cores	O
and	O
are	O
programmed	O
using	O
programming	O
models	O
such	O
as	O
CUDA	B-Architecture
or	O
OpenCL	B-Application
.	O
</s>
<s>
Moreover	O
,	O
it	O
is	O
quite	O
difficult	O
to	O
debug	O
and	O
test	O
parallel	B-Operating_System
programs	I-Operating_System
.	O
</s>
<s>
Opportunistic	O
supercomputing	B-Architecture
is	O
a	O
form	O
of	O
networked	O
grid	B-Architecture
computing	I-Architecture
whereby	O
a	O
"	O
super	O
virtual	O
computer	O
"	O
of	O
many	O
loosely	B-Application
coupled	I-Application
volunteer	B-Operating_System
computing	I-Operating_System
machines	O
performs	O
very	O
large	O
computing	O
tasks	O
.	O
</s>
<s>
Grid	B-Architecture
computing	I-Architecture
has	O
been	O
applied	O
to	O
a	O
number	O
of	O
large-scale	O
embarrassingly	B-Operating_System
parallel	I-Operating_System
problems	I-Operating_System
that	O
require	O
supercomputing	B-Architecture
performance	O
scales	O
.	O
</s>
<s>
However	O
,	O
basic	O
grid	O
and	O
cloud	B-Architecture
computing	I-Architecture
approaches	O
that	O
rely	O
on	O
volunteer	B-Operating_System
computing	I-Operating_System
cannot	O
handle	O
traditional	O
supercomputing	B-Architecture
tasks	O
such	O
as	O
fluid	O
dynamic	O
simulations	O
.	O
</s>
<s>
The	O
fastest	O
grid	B-Architecture
computing	I-Architecture
system	O
is	O
the	O
volunteer	B-Operating_System
computing	I-Operating_System
project	I-Operating_System
Folding	B-Application
@home	I-Application
(	O
F	B-Application
@h	I-Application
)	O
.	O
</s>
<s>
,	O
F	B-Application
@h	I-Application
reported	O
2.5exaFLOPS	O
of	O
x86	B-Operating_System
processing	O
power	O
.	O
</s>
<s>
The	O
Berkeley	B-Operating_System
Open	I-Operating_System
Infrastructure	I-Operating_System
for	I-Operating_System
Network	I-Operating_System
Computing	I-Operating_System
(	O
BOINC	B-Operating_System
)	O
platform	O
hosts	O
a	O
number	O
of	O
volunteer	B-Operating_System
computing	I-Operating_System
projects	O
.	O
</s>
<s>
,	O
BOINC	B-Operating_System
recorded	O
a	O
processing	O
power	O
of	O
over	O
166petaFLOPS	O
through	O
over	O
762thousand	O
active	O
Computers	O
(	O
Hosts	O
)	O
on	O
the	O
network	B-Architecture
.	O
</s>
<s>
The	O
supports	O
GIMPS	O
's	O
grid	B-Architecture
computing	I-Architecture
approach	O
,	O
one	O
of	O
the	O
earliest	O
volunteer	B-Operating_System
computing	I-Operating_System
projects	O
,	O
since	O
1997	O
.	O
</s>
<s>
Quasi-opportunistic	O
supercomputing	B-Architecture
is	O
a	O
form	O
of	O
distributed	B-Architecture
computing	I-Architecture
whereby	O
the	O
"	O
super	O
virtual	O
computer	O
"	O
of	O
many	O
networked	O
geographically	O
disperse	O
computers	O
performs	O
computing	O
tasks	O
that	O
demand	O
huge	O
processing	O
power	O
.	O
</s>
<s>
Quasi-opportunistic	O
supercomputing	B-Architecture
aims	O
to	O
provide	O
a	O
higher	O
quality	O
of	O
service	O
than	O
opportunistic	B-Architecture
grid	I-Architecture
computing	I-Architecture
by	O
achieving	O
more	O
control	O
over	O
the	O
assignment	O
of	O
tasks	O
to	O
distributed	O
resources	O
and	O
the	O
use	O
of	O
intelligence	O
about	O
the	O
availability	O
and	O
reliability	O
of	O
individual	O
systems	O
within	O
the	O
supercomputing	B-Architecture
network	B-Architecture
.	O
</s>
<s>
However	O
,	O
quasi-opportunistic	O
distributed	O
execution	O
of	O
demanding	O
parallel	B-Operating_System
computing	I-Operating_System
software	O
in	O
grids	O
should	O
be	O
achieved	O
through	O
implementation	O
of	O
grid-wise	O
allocation	O
agreements	O
,	O
co-allocation	O
subsystems	O
,	O
communication	O
topology-aware	O
allocation	O
mechanisms	O
,	O
fault	O
tolerant	O
message	O
passing	O
libraries	O
and	O
data	O
pre-conditioning	O
.	O
</s>
<s>
Cloud	B-Architecture
computing	I-Architecture
with	O
its	O
recent	O
and	O
rapid	O
expansions	O
and	O
development	O
have	O
grabbed	O
the	O
attention	O
of	O
high-performance	B-Architecture
computing	I-Architecture
(	O
HPC	O
)	O
users	O
and	O
developers	O
in	O
recent	O
years	O
.	O
</s>
<s>
Cloud	B-Architecture
computing	I-Architecture
attempts	O
to	O
provide	O
HPC-as-a-service	O
exactly	O
like	O
other	O
forms	O
of	O
services	O
available	O
in	B-Architecture
the	I-Architecture
cloud	I-Architecture
such	O
as	O
software	B-Architecture
as	I-Architecture
a	I-Architecture
service	I-Architecture
,	O
platform	B-Application
as	I-Application
a	I-Application
service	I-Application
,	O
and	O
infrastructure	B-General_Concept
as	I-General_Concept
a	I-General_Concept
service	I-General_Concept
.	O
</s>
<s>
Good	O
examples	O
of	O
such	O
challenges	O
are	O
virtualization	B-General_Concept
overhead	O
in	B-Architecture
the	I-Architecture
cloud	I-Architecture
,	O
multi-tenancy	O
of	O
resources	O
,	O
and	O
network	B-Architecture
latency	O
issues	O
.	O
</s>
<s>
Much	O
research	O
is	O
currently	O
being	O
done	O
to	O
overcome	O
these	O
challenges	O
and	O
make	O
HPC	O
in	B-Architecture
the	I-Architecture
cloud	I-Architecture
a	O
more	O
realistic	O
possibility	O
.	O
</s>
<s>
In	O
2016	O
,	O
Penguin	O
Computing	O
,	O
Parallel	O
Works	O
,	O
R-HPC	O
,	O
Amazon	B-Application
Web	I-Application
Services	I-Application
,	O
Univa	B-General_Concept
,	O
Silicon	O
Graphics	O
International	O
,	O
Rescale	B-Application
,	O
Sabalcore	O
,	O
and	O
Gomput	O
started	O
to	O
offer	O
HPC	O
cloud	B-Architecture
computing	I-Architecture
.	O
</s>
<s>
The	O
Penguin	O
On	O
Demand	O
(	O
POD	O
)	O
cloud	O
is	O
a	O
bare-metal	O
compute	O
model	O
to	O
execute	O
code	O
,	O
but	O
each	O
user	O
is	O
given	O
virtualized	B-General_Concept
login	O
node	O
.	O
</s>
<s>
POD	O
computing	O
nodes	B-Operating_System
are	O
connected	O
via	O
non-virtualized	O
10	O
Gbit/s	O
Ethernet	O
or	O
QDR	O
InfiniBand	B-Architecture
networks	O
.	O
</s>
<s>
User	O
connectivity	O
to	O
the	O
POD	O
data	B-Operating_System
center	I-Operating_System
ranges	O
from	O
50Mbit/s	O
to	O
1Gbit/s	O
.	O
</s>
<s>
Citing	O
Amazon	O
's	O
EC2	O
Elastic	O
Compute	O
Cloud	O
,	O
Penguin	O
Computing	O
argues	O
that	O
virtualization	B-General_Concept
of	O
compute	O
nodes	B-Operating_System
is	O
not	O
suitable	O
for	O
HPC	O
.	O
</s>
<s>
Penguin	O
Computing	O
has	O
also	O
criticized	O
that	O
HPC	O
clouds	O
may	O
have	O
allocated	O
computing	O
nodes	B-Operating_System
to	O
customers	O
that	O
are	O
far	O
apart	O
,	O
causing	O
latency	O
that	O
impairs	O
performance	O
for	O
some	O
HPC	O
applications	O
.	O
</s>
<s>
Supercomputers	B-Architecture
generally	O
aim	O
for	O
the	O
maximum	O
in	O
capability	O
computing	O
rather	O
than	O
capacity	O
computing	O
.	O
</s>
<s>
a	O
very	O
complex	O
weather	B-General_Concept
simulation	I-General_Concept
application	O
.	O
</s>
<s>
Architectures	O
that	O
lend	O
themselves	O
to	O
supporting	O
many	O
users	O
for	O
routine	O
everyday	O
tasks	O
may	O
have	O
a	O
lot	O
of	O
capacity	O
but	O
are	O
not	O
typically	O
considered	O
supercomputers	B-Architecture
,	O
given	O
that	O
they	O
do	O
not	O
solve	O
a	O
single	O
very	O
complex	O
problem	O
.	O
</s>
<s>
In	O
general	O
,	O
the	O
speed	O
of	O
supercomputers	B-Architecture
is	O
measured	O
and	O
benchmarked	O
in	O
FLOPS	O
(	O
floating-point	B-Algorithm
operations	O
per	O
second	O
)	O
,	O
and	O
not	O
in	O
terms	O
of	O
MIPS	O
(	O
million	O
instructions	O
per	O
second	O
)	O
,	O
as	O
is	O
the	O
case	O
with	O
general-purpose	O
computers	O
.	O
</s>
<s>
Petascale	B-General_Concept
supercomputers	B-Architecture
can	O
process	O
one	O
quadrillion	O
(	O
1015	O
)	O
(	O
1000trillion	O
)	O
FLOPS	O
.	O
</s>
<s>
Exascale	B-General_Concept
is	O
computing	O
performance	O
in	O
the	O
exaFLOPS	O
(	O
EFLOPS	O
)	O
range	O
.	O
</s>
<s>
No	O
single	O
number	O
can	O
reflect	O
the	O
overall	O
performance	O
of	O
a	O
computer	O
system	O
,	O
yet	O
the	O
goal	O
of	O
the	O
Linpack	B-Device
benchmark	I-Device
is	O
to	O
approximate	O
how	O
fast	O
the	O
computer	O
solves	O
numerical	O
problems	O
and	O
it	O
is	O
widely	O
used	O
in	O
the	O
industry	O
.	O
</s>
<s>
The	O
FLOPS	O
measurement	O
is	O
either	O
quoted	O
based	O
on	O
the	O
theoretical	O
floating	B-Algorithm
point	I-Algorithm
performance	O
of	O
a	O
processor	O
(	O
derived	O
from	O
manufacturer	O
's	O
processor	O
specifications	O
and	O
shown	O
as	O
"	O
Rpeak	O
"	O
in	O
the	O
TOP500	B-Operating_System
lists	O
)	O
,	O
which	O
is	O
generally	O
unachievable	O
when	O
running	O
real	O
workloads	O
,	O
or	O
the	O
achievable	O
throughput	O
,	O
derived	O
from	O
the	O
LINPACK	B-Device
benchmarks	I-Device
and	O
shown	O
as	O
"	O
Rmax	O
"	O
in	O
the	O
TOP500	B-Operating_System
list	I-Operating_System
.	O
</s>
<s>
The	O
LINPACK	B-Device
benchmark	I-Device
typically	O
performs	O
LU	O
decomposition	O
of	O
a	O
large	O
matrix	O
.	O
</s>
<s>
The	O
LINPACK	O
performance	O
gives	O
some	O
indication	O
of	O
performance	O
for	O
some	O
real-world	O
problems	O
,	O
but	O
does	O
not	O
necessarily	O
match	O
the	O
processing	O
requirements	O
of	O
many	O
other	O
supercomputer	B-Architecture
workloads	O
,	O
which	O
for	O
example	O
may	O
require	O
more	O
memory	O
bandwidth	O
,	O
or	O
may	O
require	O
better	O
integer	O
computing	O
performance	O
,	O
or	O
may	O
need	O
a	O
high	O
performance	O
I/O	B-General_Concept
system	O
to	O
achieve	O
high	O
levels	O
of	O
performance	O
.	O
</s>
<s>
Since	O
1993	O
,	O
the	O
fastest	B-Operating_System
supercomputers	I-Operating_System
have	O
been	O
ranked	O
on	O
the	O
TOP500	B-Operating_System
list	I-Operating_System
according	O
to	O
their	O
LINPACK	B-Device
benchmark	I-Device
results	O
.	O
</s>
<s>
The	O
list	O
does	O
not	O
claim	O
to	O
be	O
unbiased	O
or	O
definitive	O
,	O
but	O
it	O
is	O
a	O
widely	O
cited	O
current	O
definition	O
of	O
the	O
"	O
fastest	O
"	O
supercomputer	B-Architecture
available	O
at	O
any	O
given	O
time	O
.	O
</s>
<s>
This	O
is	O
a	O
recent	O
list	O
of	O
the	O
computers	O
which	O
appeared	O
at	O
the	O
top	O
of	O
the	O
TOP500	B-Operating_System
list	I-Operating_System
,	O
and	O
the	O
"	O
Peak	O
speed	O
"	O
is	O
given	O
as	O
the	O
"	O
Rmax	O
"	O
rating	O
.	O
</s>
<s>
In	O
2018	O
,	O
Lenovo	O
became	O
the	O
world	O
's	O
largest	O
provider	O
for	O
the	O
TOP500	B-Operating_System
supercomputers	B-Architecture
with	O
117	O
units	O
produced	O
.	O
</s>
<s>
(	O
2,352	O
×	O
64-core	O
Optimized	O
3rd	O
Generation	O
EPYC	O
64C	O
@2	O
.0	O
GHz	O
)	O
9,408	O
×	O
220	O
AMD	B-Operating_System
Instinct	I-Operating_System
MI250XSlingshot-11HPEEuroHPC	O
JU	O
,	O
</s>
<s>
(	O
3,456	O
×	O
32-core	O
Xeon	O
Platinum	O
8358	O
@2	O
.6	O
GHz	O
)	O
13,824	O
×	O
108	O
Nvidia	B-General_Concept
Ampere	I-General_Concept
A100Nvidia	O
HDR100	O
InfinibandAtosEuroHPC	O
JU	O
,	O
</s>
<s>
The	O
stages	O
of	O
supercomputer	B-Architecture
application	O
may	O
be	O
summarized	O
in	O
the	O
following	O
table	O
:	O
</s>
<s>
The	O
same	O
research	O
group	O
also	O
succeeded	O
in	O
using	O
a	O
supercomputer	B-Architecture
to	O
simulate	O
a	O
number	O
of	O
artificial	O
neurons	O
equivalent	O
to	O
the	O
entirety	O
of	O
a	O
rat	O
's	O
brain	O
.	O
</s>
<s>
Modern-day	O
weather	O
forecasting	O
also	O
relies	O
on	O
supercomputers	B-Architecture
.	O
</s>
<s>
The	O
National	O
Oceanic	O
and	O
Atmospheric	O
Administration	O
uses	O
supercomputers	B-Architecture
to	O
crunch	O
hundreds	O
of	O
millions	O
of	O
observations	O
to	O
help	O
make	O
weather	O
forecasts	O
more	O
accurate	O
.	O
</s>
<s>
In	O
2011	O
,	O
the	O
challenges	O
and	O
difficulties	O
in	O
pushing	O
the	O
envelope	O
in	O
supercomputing	B-Architecture
were	O
underscored	O
by	O
IBM	O
's	O
abandonment	O
of	O
the	O
Blue	B-Device
Waters	I-Device
petascale	B-General_Concept
project	O
.	O
</s>
<s>
The	O
Advanced	B-General_Concept
Simulation	I-General_Concept
and	I-General_Concept
Computing	I-General_Concept
Program	I-General_Concept
currently	O
uses	O
supercomputers	B-Architecture
to	O
maintain	O
and	O
simulate	O
the	O
United	O
States	O
nuclear	O
stockpile	O
.	O
</s>
<s>
Supercomputers	B-Architecture
used	O
different	O
simulations	O
to	O
find	O
compounds	O
that	O
could	O
potentially	O
stop	O
the	O
spread	O
.	O
</s>
<s>
In	O
the	O
2010s	O
,	O
China	O
,	O
the	O
United	O
States	O
,	O
the	O
European	O
Union	O
,	O
and	O
others	O
competed	O
to	O
be	O
the	O
first	O
to	O
create	O
a	O
1exaFLOP	O
(	O
1018	O
or	O
one	O
quintillion	O
FLOPS	O
)	O
supercomputer	B-Architecture
.	O
</s>
<s>
Many	O
Monte	B-Algorithm
Carlo	I-Algorithm
simulations	I-Algorithm
use	O
the	O
same	O
algorithm	O
to	O
process	O
a	O
randomly	O
generated	O
data	O
set	O
;	O
particularly	O
,	O
integro-differential	O
equations	O
describing	O
physical	O
transport	O
processes	O
,	O
the	O
random	O
paths	O
,	O
collisions	O
,	O
and	O
energy	O
and	O
momentum	O
depositions	O
of	O
neutrons	O
,	O
photons	O
,	O
ions	O
,	O
electrons	O
,	O
etc	O
.	O
</s>
<s>
The	O
next	O
step	O
for	O
microprocessors	B-Architecture
may	O
be	O
into	O
the	O
third	B-Architecture
dimension	I-Architecture
;	O
and	O
specializing	O
to	O
Monte	O
Carlo	O
,	O
the	O
many	O
layers	O
could	O
be	O
identical	O
,	O
simplifying	O
the	O
design	O
and	O
manufacture	O
process	O
.	O
</s>
<s>
The	O
cost	O
of	O
operating	O
high	O
performance	O
supercomputers	B-Architecture
has	O
risen	O
,	O
mainly	O
due	O
to	O
increasing	O
power	O
consumption	O
.	O
</s>
<s>
In	O
the	O
mid-1990s	O
a	O
top	O
10	O
supercomputer	B-Architecture
required	O
in	O
the	O
range	O
of	O
100kilowatts	O
,	O
in	O
2010	O
the	O
top	O
10	O
supercomputers	B-Architecture
required	O
between	O
1	O
and	O
2megawatts	O
.	O
</s>
<s>
A	O
2010	O
study	O
commissioned	O
by	O
DARPA	O
identified	O
power	O
consumption	O
as	O
the	O
most	O
pervasive	O
challenge	O
in	O
achieving	O
Exascale	B-General_Concept
computing	I-General_Concept
.	O
</s>
<s>
Supercomputing	B-Architecture
facilities	O
were	O
constructed	O
to	O
efficiently	O
remove	O
the	O
increasing	O
amount	O
of	O
heat	O
produced	O
by	O
modern	O
multi-core	B-Architecture
central	B-General_Concept
processing	I-General_Concept
units	I-General_Concept
.	O
</s>
<s>
Based	O
on	O
the	O
energy	O
consumption	O
of	O
the	O
Green	B-Application
500	I-Application
list	B-Operating_System
of	I-Operating_System
supercomputers	I-Operating_System
between	O
2007	O
and	O
2011	O
,	O
a	O
supercomputer	B-Architecture
with	O
1exaFLOPS	O
in	O
2011	O
would	O
have	O
required	O
nearly	O
500megawatts	O
.	O
</s>
<s>
Operating	B-General_Concept
systems	I-General_Concept
were	O
developed	O
for	O
existing	O
hardware	O
to	O
conserve	O
energy	O
whenever	O
possible	O
.	O
</s>
<s>
CPU	B-Architecture
cores	I-Architecture
not	O
in	O
use	O
during	O
the	O
execution	O
of	O
a	O
parallelized	B-Operating_System
application	O
were	O
put	O
into	O
low-power	O
states	O
,	O
producing	O
energy	O
savings	O
for	O
some	O
supercomputing	B-Architecture
applications	O
.	O
</s>
<s>
The	O
increasing	O
cost	O
of	O
operating	O
supercomputers	B-Architecture
has	O
been	O
a	O
driving	O
factor	O
in	O
a	O
trend	O
toward	O
bundling	O
of	O
resources	O
through	O
a	O
distributed	O
supercomputer	B-Architecture
infrastructure	O
.	O
</s>
<s>
National	B-Device
supercomputing	I-Device
centers	I-Device
first	O
emerged	O
in	O
the	O
US	O
,	O
followed	O
by	O
Germany	O
and	O
Japan	O
.	O
</s>
<s>
The	O
European	O
Union	O
launched	O
the	O
Partnership	O
for	O
Advanced	O
Computing	O
in	O
Europe	O
(	O
PRACE	O
)	O
with	O
the	O
aim	O
of	O
creating	O
a	O
persistent	O
pan-European	O
supercomputer	B-Architecture
infrastructure	O
with	O
services	O
to	O
support	O
scientists	O
across	O
the	O
European	O
Union	O
in	O
porting	O
,	O
scaling	O
and	O
optimizing	O
supercomputing	B-Architecture
applications	O
.	O
</s>
<s>
Iceland	O
built	O
the	O
world	O
's	O
first	O
zero-emission	O
supercomputer	B-Architecture
.	O
</s>
<s>
Located	O
at	O
the	O
Thor	O
Data	B-Operating_System
Center	I-Operating_System
in	O
Reykjavík	O
,	O
Iceland	O
,	O
this	O
supercomputer	B-Architecture
relies	O
on	O
completely	O
renewable	O
sources	O
for	O
its	O
power	O
rather	O
than	O
fossil	O
fuels	O
.	O
</s>
<s>
Funding	O
supercomputer	B-Architecture
hardware	O
also	O
became	O
increasingly	O
difficult	O
.	O
</s>
<s>
In	O
the	O
mid-1990s	O
a	O
top	O
10	O
supercomputer	B-Architecture
cost	O
about	O
10million	O
euros	O
,	O
while	O
in	O
2010	O
the	O
top	O
10	O
supercomputers	B-Architecture
required	O
an	O
investment	O
of	O
between	O
40	O
and	O
50million	O
euros	O
.	O
</s>
<s>
In	O
the	O
2000s	O
national	O
governments	O
put	O
in	O
place	O
different	O
strategies	O
to	O
fund	O
supercomputers	B-Architecture
.	O
</s>
<s>
In	O
the	O
UK	O
the	O
national	O
government	O
funded	O
supercomputers	B-Architecture
entirely	O
and	O
high	B-Architecture
performance	I-Architecture
computing	I-Architecture
was	O
put	O
under	O
the	O
control	O
of	O
a	O
national	O
funding	O
agency	O
.	O
</s>
<s>
Many	O
science	O
fiction	O
writers	O
have	O
depicted	O
supercomputers	B-Architecture
in	O
their	O
works	O
,	O
both	O
before	O
and	O
after	O
the	O
historical	O
construction	O
of	O
such	O
computers	O
.	O
</s>
<s>
Examples	O
of	O
supercomputers	B-Architecture
in	O
fiction	O
include	O
HAL	B-General_Concept
9000	I-General_Concept
,	O
Multivac	O
,	O
The	O
Machine	O
Stops	O
,	O
GLaDOS	O
,	O
The	O
Evitable	O
Conflict	O
,	O
Vulcan	O
's	O
Hammer	O
,	O
Colossus	O
,	O
WOPR	B-Application
,	O
and	O
Deep	O
Thought	O
.	O
</s>
