<s>
Summit	B-Device
or	O
OLCF-4	B-Device
is	O
a	O
supercomputer	B-Architecture
developed	O
by	O
IBM	O
for	O
use	O
at	O
Oak	B-Device
Ridge	I-Device
Leadership	I-Device
Computing	I-Device
Facility	I-Device
(	O
OLCF	B-Device
)	O
,	O
a	O
facility	O
at	O
the	O
Oak	O
Ridge	O
National	O
Laboratory	O
,	O
capable	O
of	O
200	O
petaFLOPS	O
thus	O
making	B-Operating_System
it	I-Operating_System
the	I-Operating_System
5th	I-Operating_System
fastest	I-Operating_System
supercomputer	I-Operating_System
in	I-Operating_System
the	I-Operating_System
world	I-Operating_System
after	O
Frontier	B-Device
(	O
OLCF-5	B-Device
)	O
,	O
Fugaku	B-Device
,	O
LUMI	B-Device
and	O
Leonardo	O
.	O
</s>
<s>
Its	O
current	O
LINPACK	B-Device
benchmark	I-Device
is	O
clocked	O
at	O
148.6	O
petaFLOPS	O
.	O
</s>
<s>
As	O
of	O
November	O
2019	O
,	O
the	O
supercomputer	B-Architecture
had	O
ranked	O
as	O
the	O
5th	O
most	O
energy	O
efficient	O
in	O
the	O
world	O
with	O
a	O
measured	O
power	O
efficiency	O
of	O
14.668	O
gigaFLOPS/watt	O
.	O
</s>
<s>
Summit	B-Device
was	O
the	O
first	O
supercomputer	B-Architecture
to	O
reach	O
exaflop	O
(	O
a	O
quintillion	O
operations	O
per	O
second	O
)	O
speed	O
,	O
achieving	O
1.88	O
exaflops	O
during	O
a	O
genomic	O
analysis	O
and	O
is	O
expected	O
to	O
reach	O
3.3	O
exaflops	O
using	O
mixed-precision	B-Algorithm
calculations	O
.	O
</s>
<s>
The	O
effort	O
resulted	O
in	O
construction	O
of	O
Summit	B-Device
and	O
Sierra	B-Device
.	O
</s>
<s>
Summit	B-Device
is	O
tasked	O
with	O
civilian	O
scientific	O
research	O
and	O
is	O
located	O
at	O
the	O
Oak	O
Ridge	O
National	O
Laboratory	O
in	O
Tennessee	O
.	O
</s>
<s>
Sierra	B-Device
is	O
designed	O
for	O
nuclear	O
weapons	O
simulations	O
and	O
is	O
located	O
at	O
the	O
Lawrence	O
Livermore	O
National	O
Laboratory	O
in	O
California	O
.	O
</s>
<s>
Summit	B-Device
is	O
estimated	O
to	O
cover	O
873	O
square	O
meters	O
and	O
require	O
219	O
kilometers	O
of	O
cabling	O
.	O
</s>
<s>
Researchers	O
will	O
utilize	O
Summit	B-Device
for	O
diverse	O
fields	O
such	O
as	O
cosmology	O
,	O
medicine	O
,	O
and	O
climatology	O
.	O
</s>
<s>
In	O
2015	O
,	O
the	O
project	O
called	O
Collaboration	O
of	O
Oak	O
Ridge	O
,	O
Argonne	O
and	O
Lawrence	O
Livermore	O
(	O
CORAL	O
)	O
included	O
a	O
third	O
supercomputer	B-Architecture
named	O
Aurora	B-Device
and	O
was	O
planned	O
for	O
installation	O
at	O
Argonne	O
National	O
Laboratory	O
.	O
</s>
<s>
By	O
2018	O
,	O
Aurora	B-Device
was	O
re-engineered	O
with	O
completion	O
anticipated	O
in	O
2021	O
as	O
an	O
exascale	B-General_Concept
computing	I-General_Concept
project	O
along	O
with	O
Frontier	B-Device
and	O
El	O
Capitan	O
to	O
be	O
completed	O
shortly	O
thereafter	O
;	O
Aurora	B-Device
was	O
completed	O
in	O
late	O
2022	O
.	O
</s>
<s>
The	O
Summit	B-Device
supercomputer	I-Device
provides	O
scientists	O
and	O
researchers	O
the	O
opportunity	O
to	O
solve	O
complex	O
tasks	O
in	O
the	O
fields	O
of	O
energy	O
,	O
artificial	O
intelligence	O
,	O
human	O
health	O
and	O
other	O
research	O
areas	O
.	O
</s>
<s>
Each	O
one	O
of	O
its	O
4,608	O
nodes	O
(	O
with	O
2	O
IBM	O
POWER9	B-Device
CPUs	B-Device
and	O
6	O
Nvidia	B-Device
Tesla	I-Device
GPUs	B-Architecture
in	O
each	O
node	O
)	O
has	O
over	O
600	O
GB	O
of	O
coherent	B-Operating_System
memory	I-Operating_System
(	O
96	O
GB	O
HBM2	O
plus	O
512	O
GB	O
DDR4	O
SDRAM	O
)	O
which	O
is	O
addressable	O
by	O
all	O
CPUs	B-Device
and	O
GPUs	B-Architecture
plus	O
800	O
GB	O
of	O
non-volatile	B-General_Concept
RAM	I-General_Concept
that	O
can	O
be	O
used	O
as	O
a	O
burst	B-Operating_System
buffer	I-Operating_System
or	O
as	O
extended	O
memory	O
.	O
</s>
<s>
The	O
POWER9	B-Device
CPUs	B-Device
and	O
Nvidia	B-General_Concept
Volta	I-General_Concept
GPUs	B-Architecture
are	O
connected	O
using	O
NVIDIA	O
's	O
high	O
speed	O
NVLink	O
.	O
</s>
<s>
To	O
provide	O
a	O
high	O
rate	O
of	O
data	O
throughput	O
,	O
the	O
nodes	O
will	O
be	O
connected	O
in	O
a	O
non-blocking	O
fat-tree	B-Protocol
topology	I-Protocol
using	O
a	O
dual-rail	O
Mellanox	O
EDR	O
InfiniBand	B-Architecture
interconnect	O
for	O
both	O
storage	O
and	O
inter-process	O
communications	O
traffic	O
which	O
delivers	O
both	O
200Gbit/s	O
bandwidth	O
between	O
nodes	O
and	O
in-network	O
computing	O
acceleration	O
for	O
communications	O
frameworks	O
such	O
as	O
MPI	B-Application
and	O
SHMEM/PGAS	O
.	O
</s>
