<s>
The	O
is	O
a	O
series	O
of	O
supercomputers	B-Architecture
deployed	O
at	O
Japan	O
Agency	O
for	O
Marine-Earth	O
Science	O
and	O
Technology	O
Yokohama	O
Institute	O
of	O
Earth	O
Sciences	O
.	O
</s>
<s>
The	O
first	O
generation	O
of	O
Earth	B-Device
Simulator	I-Device
,	O
developed	O
by	O
the	O
Japanese	O
government	O
's	O
initiative	O
"	O
Earth	B-Device
Simulator	I-Device
Project	O
"	O
,	O
was	O
a	O
highly	O
parallel	O
vector	O
supercomputer	B-Architecture
system	O
for	O
running	O
global	B-Application
climate	I-Application
models	I-Application
to	O
evaluate	O
the	O
effects	O
of	O
global	O
warming	O
and	O
problems	O
in	O
solid	O
earth	O
geophysics	O
.	O
</s>
<s>
Built	O
by	O
NEC	O
,	O
ES	O
was	O
based	O
on	O
their	O
SX-6	B-Device
architecture	O
.	O
</s>
<s>
It	O
consisted	O
of	O
640	O
nodes	O
with	O
eight	O
vector	B-Operating_System
processors	I-Operating_System
and	O
16	O
gigabytes	O
of	O
computer	B-General_Concept
memory	I-General_Concept
at	O
each	O
node	O
,	O
for	O
a	O
total	O
of	O
5120	O
processors	O
and	O
10	O
terabytes	O
of	O
memory	O
.	O
</s>
<s>
The	O
system	O
had	O
700	O
terabytes	O
of	O
disk	B-Device
storage	I-Device
(	O
450	O
for	O
the	O
system	O
and	O
250	O
for	O
the	O
users	O
)	O
and	O
1.6	O
petabytes	O
of	O
mass	B-Device
storage	I-Device
in	O
tape	O
drives	O
.	O
</s>
<s>
Its	O
performance	O
on	O
the	O
LINPACK	B-Application
benchmark	O
was	O
35.86	O
TFLOPS	O
,	O
which	O
was	O
almost	O
five	O
times	O
faster	O
than	O
the	O
previous	O
fastest	B-Operating_System
supercomputer	I-Operating_System
,	O
ASCI	B-Operating_System
White	I-Operating_System
.	O
</s>
<s>
As	O
of	O
2020	O
,	O
comparable	O
performance	O
can	O
be	O
achieved	O
by	O
using	O
4	O
Nvidia	B-General_Concept
A100	I-General_Concept
GPUs	O
,	O
each	O
with	O
9.746	O
FP64	O
TFlops	O
.	O
</s>
<s>
ES	O
was	O
the	O
fastest	B-Operating_System
supercomputer	I-Operating_System
in	I-Operating_System
the	I-Operating_System
world	I-Operating_System
from	O
2002	O
to	O
2004	O
.	O
</s>
<s>
Its	O
capacity	O
was	O
surpassed	O
by	O
IBM	O
's	O
Blue	B-Operating_System
Gene/L	I-Operating_System
prototype	O
on	O
29	O
September	O
2004	O
.	O
</s>
<s>
ES	O
was	O
replaced	O
by	O
the	O
Earth	B-Device
Simulator	I-Device
2	O
(	O
ES2	O
)	O
in	O
March	O
2009	O
.	O
</s>
<s>
With	O
a	O
delivered	O
LINPACK	B-Application
performance	O
of	O
122.4	O
TFLOPS	O
,	O
ES2	O
was	O
the	O
most	O
efficient	O
supercomputer	B-Architecture
in	O
the	O
world	O
at	O
that	O
point	O
.	O
</s>
<s>
In	O
November	O
2010	O
,	O
NEC	O
announced	O
that	O
ES2	O
topped	O
the	O
Global	O
FFT	O
,	O
one	O
of	O
the	O
measures	O
of	O
the	O
HPC	B-Application
Challenge	I-Application
Awards	I-Application
,	O
with	O
the	O
performance	O
number	O
of	O
11.876	O
TFLOPS	O
.	O
</s>
<s>
ES2	O
was	O
replaced	O
by	O
the	O
Earth	B-Device
Simulator	I-Device
3	O
(	O
ES3	O
)	O
in	O
March	O
2015	O
.	O
</s>
<s>
ES3	O
is	O
a	O
NEC	B-Device
SX-ACE	I-Device
system	O
with	O
5120	O
nodes	O
,	O
and	O
a	O
performance	O
of	O
1.3	O
PFLOPS	O
.	O
</s>
<s>
ES3	O
,	O
from	O
2017	O
to	O
2018	O
,	O
ran	O
alongside	O
Gyoukou	O
,	O
a	O
supercomputer	B-Architecture
with	O
immersion	O
cooling	O
that	O
can	O
achieve	O
up	O
to	O
19	O
PFLOPS	O
.	O
</s>
<s>
The	O
Earth	B-Device
Simulator	I-Device
4	O
(	O
ES4	O
)	O
uses	O
AMD	O
EPYC	O
processors	O
,	O
with	O
acceleration	O
by	O
the	O
NEC	B-Device
SX-Aurora	I-Device
TSUBASA	I-Device
Vector	O
Engine	O
and	O
NVIDIA	B-General_Concept
Ampere	I-General_Concept
A100	I-General_Concept
GPUs	O
.	O
</s>
