<s>
Blue	B-Operating_System
Gene	I-Operating_System
is	O
an	O
IBM	O
project	O
aimed	O
at	O
designing	O
supercomputers	B-Architecture
that	O
can	O
reach	O
operating	O
speeds	O
in	O
the	O
petaFLOPS	O
(	O
PFLOPS	O
)	O
range	O
,	O
with	O
low	O
power	O
consumption	O
.	O
</s>
<s>
The	O
project	O
created	O
three	O
generations	O
of	O
supercomputers	B-Architecture
,	O
Blue	B-Operating_System
Gene/L	I-Operating_System
,	O
Blue	B-Operating_System
Gene/P	I-Operating_System
,	O
and	O
Blue	O
Gene/Q	O
.	O
</s>
<s>
During	O
their	O
deployment	O
,	O
Blue	B-Operating_System
Gene	I-Operating_System
systems	O
often	O
led	O
the	O
TOP500	B-Operating_System
and	O
Green500	B-Application
rankings	O
of	O
the	O
most	O
powerful	O
and	O
most	O
power-efficient	O
supercomputers	B-Architecture
,	O
respectively	O
.	O
</s>
<s>
Blue	B-Operating_System
Gene	I-Operating_System
systems	O
have	O
also	O
consistently	O
scored	O
top	O
positions	O
in	O
the	O
Graph500	B-Device
list	O
.	O
</s>
<s>
As	O
of	O
2015	O
,	O
IBM	O
appears	O
to	O
have	O
ended	O
development	O
of	O
the	O
Blue	B-Operating_System
Gene	I-Operating_System
family	O
,	O
though	O
no	O
formal	O
announcement	O
has	O
been	O
made	O
.	O
</s>
<s>
IBM	O
has	O
since	O
focused	O
its	O
supercomputer	B-Architecture
efforts	O
on	O
the	O
OpenPower	B-Application
platform	O
,	O
using	O
accelerators	O
such	O
as	O
FPGAs	B-Architecture
and	O
GPUs	B-Architecture
to	O
address	O
the	O
diminishing	O
returns	O
of	O
Moore	O
's	O
law	O
.	O
</s>
<s>
In	O
December	O
1999	O
,	O
IBM	O
announced	O
a	O
US$100	O
million	O
research	O
initiative	O
for	O
a	O
five-year	O
effort	O
to	O
build	O
a	O
massively	O
parallel	B-Operating_System
computer	I-Operating_System
,	O
to	O
be	O
applied	O
to	O
the	O
study	O
of	O
biomolecular	O
phenomena	O
such	O
as	O
protein	O
folding	O
.	O
</s>
<s>
The	O
project	O
had	O
two	O
main	O
goals	O
:	O
to	O
advance	O
our	O
understanding	O
of	O
the	O
mechanisms	O
behind	O
protein	O
folding	O
via	O
large-scale	O
simulation	O
,	O
and	O
to	O
explore	O
novel	O
ideas	O
in	O
massively	O
parallel	B-Operating_System
machine	I-Operating_System
architecture	O
and	O
software	O
.	O
</s>
<s>
Major	O
areas	O
of	O
investigation	O
included	O
:	O
how	O
to	O
use	O
this	O
novel	O
platform	O
to	O
effectively	O
meet	O
its	O
scientific	O
goals	O
,	O
how	O
to	O
make	O
such	O
massively	O
parallel	B-Operating_System
machines	I-Operating_System
more	O
usable	O
,	O
and	O
how	O
to	O
achieve	O
performance	O
targets	O
at	O
a	O
reasonable	O
cost	O
,	O
through	O
novel	O
machine	O
architectures	O
.	O
</s>
<s>
The	O
initial	O
design	O
for	O
Blue	B-Operating_System
Gene	I-Operating_System
was	O
based	O
on	O
an	O
early	O
version	O
of	O
the	O
Cyclops64	B-Device
architecture	O
,	O
designed	O
by	O
Monty	O
Denneau	O
.	O
</s>
<s>
At	O
IBM	O
,	O
Alan	O
Gara	O
started	O
working	O
on	O
an	O
extension	O
of	O
the	O
QCDOC	B-Operating_System
architecture	O
into	O
a	O
more	O
general-purpose	O
supercomputer	B-Architecture
:	O
The	O
4D	O
nearest-neighbor	O
interconnection	O
network	O
was	O
replaced	O
by	O
a	O
network	O
supporting	O
routing	O
of	O
messages	O
from	O
any	O
node	O
to	O
any	O
other	O
;	O
and	O
a	O
parallel	O
I/O	O
subsystem	O
was	O
added	O
.	O
</s>
<s>
DOE	O
started	O
funding	O
the	O
development	O
of	O
this	O
system	O
and	O
it	O
became	O
known	O
as	O
Blue	B-Operating_System
Gene/L	I-Operating_System
(	O
L	O
for	O
Light	O
)	O
;	O
development	O
of	O
the	O
original	O
Blue	B-Operating_System
Gene	I-Operating_System
system	O
continued	O
under	O
the	O
name	O
Blue	B-Device
Gene/C	I-Device
(	O
C	B-Language
for	O
Cyclops	O
)	O
and	O
,	O
later	O
,	O
Cyclops64	B-Device
.	O
</s>
<s>
In	O
November	O
2004	O
a	O
16-rack	O
system	O
,	O
with	O
each	O
rack	B-Application
holding	O
1,024	O
compute	O
nodes	O
,	O
achieved	O
first	O
place	O
in	O
the	O
TOP500	B-Operating_System
list	O
,	O
with	O
a	O
Linpack	B-Application
performance	O
of	O
70.72TFLOPS	O
.	O
</s>
<s>
It	O
thereby	O
overtook	O
NEC	O
's	O
Earth	B-Device
Simulator	I-Device
,	O
which	O
had	O
held	O
the	O
title	O
of	O
the	O
fastest	B-Operating_System
computer	I-Operating_System
in	O
the	O
world	O
since	O
2002	O
.	O
</s>
<s>
From	O
2004	O
through	O
2007	O
the	O
Blue	B-Operating_System
Gene/L	I-Operating_System
installation	O
at	O
LLNL	O
gradually	O
expanded	O
to	O
104	O
racks	O
,	O
achieving	O
478TFLOPS	O
Linpack	B-Application
and	O
596TFLOPS	O
peak	O
.	O
</s>
<s>
The	O
LLNL	O
BlueGene/L	B-Operating_System
installation	O
held	O
the	O
first	O
position	O
in	O
the	O
TOP500	B-Operating_System
list	O
for	O
3.5	O
years	O
,	O
until	O
in	O
June	O
2008	O
it	O
was	O
overtaken	O
by	O
IBM	O
's	O
Cell-based	O
Roadrunner	B-General_Concept
system	O
at	O
Los	O
Alamos	O
National	O
Laboratory	O
,	O
which	O
was	O
the	O
first	O
system	O
to	O
surpass	O
the	O
1	O
PetaFLOPS	O
mark	O
.	O
</s>
<s>
While	O
the	O
LLNL	O
installation	O
was	O
the	O
largest	O
Blue	B-Operating_System
Gene/L	I-Operating_System
installation	O
,	O
many	O
smaller	O
installations	O
followed	O
.	O
</s>
<s>
In	O
November	O
2006	O
,	O
there	O
were	O
27	O
computers	O
on	O
the	O
TOP500	B-Operating_System
list	O
using	O
the	O
Blue	B-Operating_System
Gene/L	I-Operating_System
architecture	O
.	O
</s>
<s>
All	O
these	O
computers	O
were	O
listed	O
as	O
having	O
an	O
architecture	O
of	O
eServer	O
Blue	B-Operating_System
Gene	I-Operating_System
Solution	O
.	O
</s>
<s>
For	O
example	O
,	O
three	O
racks	O
of	O
Blue	B-Operating_System
Gene/L	I-Operating_System
were	O
housed	O
at	O
the	O
San	O
Diego	O
Supercomputer	B-Architecture
Center	O
.	O
</s>
<s>
While	O
the	O
TOP500	B-Operating_System
measures	O
performance	O
on	O
a	O
single	O
benchmark	O
application	O
,	O
Linpack	B-Application
,	O
Blue	B-Operating_System
Gene/L	I-Operating_System
also	O
set	O
records	O
for	O
performance	O
on	O
a	O
wider	O
set	O
of	O
applications	O
.	O
</s>
<s>
Blue	B-Operating_System
Gene/L	I-Operating_System
was	O
the	O
first	O
supercomputer	B-Architecture
ever	O
to	O
run	O
over	O
100TFLOPS	O
sustained	O
on	O
a	O
real-world	O
application	O
,	O
namely	O
a	O
three-dimensional	O
molecular	O
dynamics	O
code	O
(	O
ddcMD	O
)	O
,	O
simulating	O
solidification	O
(	O
nucleation	O
and	O
growth	O
processes	O
)	O
of	O
molten	O
metal	O
under	O
high	O
pressure	O
and	O
temperature	O
conditions	O
.	O
</s>
<s>
In	O
June	O
2006	O
,	O
NNSA	O
and	O
IBM	O
announced	O
that	O
Blue	B-Operating_System
Gene/L	I-Operating_System
achieved	O
207.3TFLOPS	O
on	O
a	O
quantum	O
chemical	O
application	O
(	O
Qbox	B-Application
)	O
.	O
</s>
<s>
At	O
Supercomputing	B-Architecture
2006	O
,	O
Blue	B-Operating_System
Gene/L	I-Operating_System
was	O
awarded	O
the	O
winning	O
prize	O
in	O
all	O
HPC	O
Challenge	O
Classes	O
of	O
awards	O
.	O
</s>
<s>
In	O
2007	O
,	O
a	O
team	O
from	O
the	O
IBM	O
Almaden	O
Research	O
Center	O
and	O
the	O
University	O
of	O
Nevada	O
ran	O
an	O
artificial	B-Architecture
neural	I-Architecture
network	I-Architecture
almost	O
half	O
as	O
complex	O
as	O
the	O
brain	O
of	O
a	O
mouse	O
for	O
the	O
equivalent	O
of	O
a	O
second	O
(	O
the	O
network	O
was	O
run	O
at	O
1/10	O
of	O
normal	O
speed	O
for	O
10	O
seconds	O
)	O
.	O
</s>
<s>
The	O
name	O
Blue	B-Operating_System
Gene	I-Operating_System
comes	O
from	O
what	O
it	O
was	O
originally	O
designed	O
to	O
do	O
,	O
help	O
biologists	O
understand	O
the	O
processes	O
of	O
protein	O
folding	O
and	O
gene	O
development	O
.	O
</s>
<s>
The	O
original	O
Blue	B-Operating_System
Gene	I-Operating_System
design	O
was	O
renamed	O
"	O
Blue	B-Device
Gene/C	I-Device
"	O
and	O
eventually	O
Cyclops64	B-Device
.	O
</s>
<s>
The	O
"	O
L	O
"	O
in	O
Blue	B-Operating_System
Gene/L	I-Operating_System
comes	O
from	O
"	O
Light	O
"	O
as	O
that	O
design	O
's	O
original	O
name	O
was	O
"	O
Blue	O
Light	O
"	O
.	O
</s>
<s>
The	O
"	O
P	O
"	O
version	O
was	O
designed	O
to	O
be	O
a	O
petascale	B-General_Concept
design	O
.	O
</s>
<s>
The	O
Blue	B-Operating_System
Gene/L	I-Operating_System
supercomputer	B-Architecture
was	O
unique	O
in	O
the	O
following	O
aspects	O
:	O
</s>
<s>
Blue	B-Operating_System
Gene/L	I-Operating_System
used	O
low	O
frequency	O
and	O
low	O
power	O
embedded	O
PowerPC	O
cores	O
with	O
floating-point	B-Algorithm
accelerators	O
.	O
</s>
<s>
The	O
Blue	B-Operating_System
Gene/L	I-Operating_System
architecture	O
was	O
an	O
evolution	O
of	O
the	O
QCDSP	O
and	O
QCDOC	B-Operating_System
architectures	O
.	O
</s>
<s>
Each	O
Blue	B-Operating_System
Gene/L	I-Operating_System
Compute	O
or	O
I/O	O
node	O
was	O
a	O
single	O
ASIC	O
with	O
associated	O
DRAM	O
memory	O
chips	O
.	O
</s>
<s>
The	O
ASIC	O
integrated	O
two	O
700MHz	O
PowerPC	O
440	O
embedded	O
processors	O
,	O
each	O
with	O
a	O
double-pipeline-double-precision	O
Floating-Point	B-General_Concept
Unit	I-General_Concept
(	O
FPU	O
)	O
,	O
a	O
cache	B-General_Concept
sub-system	O
with	O
built-in	O
DRAM	O
controller	O
and	O
the	O
logic	O
to	O
support	O
multiple	O
communication	O
sub-systems	O
.	O
</s>
<s>
The	O
dual	O
FPUs	O
gave	O
each	O
Blue	B-Operating_System
Gene/L	I-Operating_System
node	O
a	O
theoretical	O
peak	O
performance	O
of	O
5.6GFLOPS	O
(	O
gigaFLOPS	O
)	O
.	O
</s>
<s>
The	O
two	O
CPUs	O
were	O
not	O
cache	B-General_Concept
coherent	I-General_Concept
with	O
one	O
another	O
.	O
</s>
<s>
There	O
were	O
32	O
node	O
boards	O
per	O
cabinet/rack	O
.	O
</s>
<s>
This	O
allowed	O
aggressive	O
packaging	O
of	O
up	O
to	O
1024	O
compute	O
nodes	O
,	O
plus	O
additional	O
I/O	O
nodes	O
,	O
in	O
a	O
standard	O
19-inch	B-Application
rack	I-Application
,	O
within	O
reasonable	O
limits	O
of	O
electrical	O
power	O
supply	O
and	O
air	O
cooling	O
.	O
</s>
<s>
The	O
system	O
was	O
able	O
to	O
electrically	O
isolate	O
faulty	O
components	O
,	O
down	O
to	O
a	O
granularity	O
of	O
half	O
a	O
rack	B-Application
(	O
512	O
compute	O
nodes	O
)	O
,	O
to	O
allow	O
the	O
machine	O
to	O
continue	O
to	O
run	O
.	O
</s>
<s>
Each	O
Blue	B-Operating_System
Gene/L	I-Operating_System
node	O
was	O
attached	O
to	O
three	O
parallel	O
communications	O
networks	O
:	O
a	O
3D	O
toroidal	B-Operating_System
network	I-Operating_System
for	O
peer-to-peer	O
communication	O
between	O
compute	O
nodes	O
,	O
a	O
collective	O
network	O
for	O
collective	O
communication	O
(	O
broadcasts	O
and	O
reduce	O
operations	O
)	O
,	O
and	O
a	O
global	O
interrupt	B-Application
network	O
for	O
fast	B-Operating_System
barriers	I-Operating_System
.	O
</s>
<s>
The	O
I/O	O
nodes	O
,	O
which	O
run	O
the	O
Linux	B-Application
operating	I-Application
system	I-Application
,	O
provided	O
communication	O
to	O
storage	O
and	O
external	O
hosts	O
via	O
an	O
Ethernet	O
network	O
.	O
</s>
<s>
Finally	O
,	O
a	O
separate	O
and	O
private	O
Ethernet	O
network	O
provided	O
access	O
to	O
any	O
node	O
for	O
configuration	O
,	O
booting	B-Operating_System
and	O
diagnostics	O
.	O
</s>
<s>
To	O
allow	O
multiple	O
programs	O
to	O
run	O
concurrently	O
,	O
a	O
Blue	B-Operating_System
Gene/L	I-Operating_System
system	O
could	O
be	O
partitioned	O
into	O
electronically	O
isolated	O
sets	O
of	O
nodes	O
.	O
</s>
<s>
To	O
run	O
a	O
program	O
on	O
Blue	B-Operating_System
Gene/L	I-Operating_System
,	O
a	O
partition	O
of	O
the	O
computer	O
was	O
first	O
to	O
be	O
reserved	O
.	O
</s>
<s>
Blue	B-Operating_System
Gene/L	I-Operating_System
compute	O
nodes	O
used	O
a	O
minimal	O
operating	B-General_Concept
system	I-General_Concept
supporting	O
a	O
single	O
user	O
program	O
.	O
</s>
<s>
Programmers	O
needed	O
to	O
implement	O
green	B-Operating_System
threads	I-Operating_System
in	O
order	O
to	O
simulate	O
local	O
concurrency	O
.	O
</s>
<s>
Application	O
development	O
was	O
usually	O
performed	O
in	O
C	B-Language
,	O
C++	B-Language
,	O
or	O
Fortran	B-Application
using	O
MPI	B-Application
for	O
communication	O
.	O
</s>
<s>
However	O
,	O
some	O
scripting	O
languages	O
such	O
as	O
Ruby	B-Language
and	O
Python	B-Language
have	O
been	O
ported	O
to	O
the	O
compute	O
nodes	O
.	O
</s>
<s>
IBM	O
published	O
BlueMatter	O
,	O
the	O
application	O
developed	O
to	O
exercise	O
Blue	B-Operating_System
Gene/L	I-Operating_System
,	O
as	O
open	O
source	O
here	O
.	O
</s>
<s>
This	O
serves	O
to	O
document	O
how	O
the	O
torus	O
and	O
collective	O
interfaces	O
were	O
used	O
by	O
applications	O
,	O
and	O
may	O
serve	O
as	O
a	O
base	O
for	O
others	O
to	O
exercise	O
the	O
current	O
generation	O
of	O
supercomputers	B-Architecture
.	O
</s>
<s>
In	O
June	O
2007	O
,	O
IBM	O
unveiled	O
Blue	B-Operating_System
Gene/P	I-Operating_System
,	O
the	O
second	O
generation	O
of	O
the	O
Blue	B-Operating_System
Gene	I-Operating_System
series	O
of	O
supercomputers	B-Architecture
and	O
designed	O
through	O
a	O
collaboration	O
that	O
included	O
IBM	O
,	O
LLNL	O
,	O
and	O
Argonne	O
National	O
Laboratory	O
's	O
.	O
</s>
<s>
The	O
design	O
of	O
Blue	B-Operating_System
Gene/P	I-Operating_System
is	O
a	O
technology	O
evolution	O
from	O
Blue	B-Operating_System
Gene/L	I-Operating_System
.	O
</s>
<s>
Each	O
Blue	B-Operating_System
Gene/P	I-Operating_System
Compute	O
chip	O
contains	O
four	O
PowerPC	O
450	O
processor	O
cores	O
,	O
running	O
at	O
850MHz	O
.	O
</s>
<s>
The	O
cores	O
are	O
cache	B-General_Concept
coherent	I-General_Concept
and	O
the	O
chip	O
can	O
operate	O
as	O
a	O
4-way	O
symmetric	B-Operating_System
multiprocessor	I-Operating_System
(	O
SMP	O
)	O
.	O
</s>
<s>
The	O
memory	O
subsystem	O
on	O
the	O
chip	O
consists	O
of	O
small	O
private	O
L2	O
caches	O
,	O
a	O
central	O
shared	O
8MB	O
L3	O
cache	B-General_Concept
,	O
and	O
dual	O
DDR2	O
memory	O
controllers	O
.	O
</s>
<s>
The	O
chip	O
also	O
integrates	O
the	O
logic	O
for	O
node-to-node	O
communication	O
,	O
using	O
the	O
same	O
network	O
topologies	O
as	O
Blue	B-Operating_System
Gene/L	I-Operating_System
,	O
but	O
at	O
more	O
than	O
twice	O
the	O
bandwidth	O
.	O
</s>
<s>
A	O
compute	O
card	O
contains	O
a	O
Blue	B-Operating_System
Gene/P	I-Operating_System
chip	O
with	O
2	O
or	O
4GB	O
DRAM	O
,	O
comprising	O
a	O
"	O
compute	O
node	O
"	O
.	O
</s>
<s>
A	O
rack	B-Application
contains	O
32	O
node	O
boards	O
(	O
thus	O
1024	O
nodes	O
,	O
4096	O
processor	O
cores	O
)	O
.	O
</s>
<s>
By	O
using	O
many	O
small	O
,	O
low-power	O
,	O
densely	O
packaged	O
chips	O
,	O
Blue	B-Operating_System
Gene/P	I-Operating_System
exceeded	O
the	O
power	O
efficiency	O
of	O
other	O
supercomputers	B-Architecture
of	O
its	O
generation	O
,	O
and	O
at	O
371MFLOPS/W	O
Blue	B-Operating_System
Gene/P	I-Operating_System
installations	O
ranked	O
at	O
or	O
near	O
the	O
top	O
of	O
the	O
Green500	B-Application
lists	O
in	O
2007-2008	O
.	O
</s>
<s>
The	O
following	O
is	O
an	O
incomplete	O
list	O
of	O
Blue	B-Operating_System
Gene/P	I-Operating_System
installations	O
.	O
</s>
<s>
Per	O
November	O
2009	O
,	O
the	O
TOP500	B-Operating_System
list	O
contained	O
15	O
Blue	B-Operating_System
Gene/P	I-Operating_System
installations	O
of	O
2-racks	O
(	O
2048	O
nodes	O
,	O
8192	O
processor	O
cores	O
,	O
23.86TFLOPS	O
Linpack	B-Application
)	O
and	O
larger	O
.	O
</s>
<s>
On	O
November	O
12	O
,	O
2007	O
,	O
the	O
first	O
Blue	B-Operating_System
Gene/P	I-Operating_System
installation	O
,	O
JUGENE	B-Operating_System
,	O
with	O
16	O
racks	O
(	O
16,384	O
nodes	O
,	O
65,536	O
processors	O
)	O
was	O
running	O
at	O
Forschungszentrum	O
Jülich	O
in	O
Germany	O
with	O
a	O
performance	O
of	O
167TFLOPS	O
.	O
</s>
<s>
When	O
inaugurated	O
it	O
was	O
the	O
fastest	B-Operating_System
supercomputer	I-Operating_System
in	O
Europe	O
and	O
the	O
sixth	O
fastest	O
in	O
the	O
world	O
.	O
</s>
<s>
In	O
2009	O
,	O
JUGENE	B-Operating_System
was	O
upgraded	O
to	O
72	O
racks	O
(	O
73,728	O
nodes	O
,	O
294,912	O
processor	O
cores	O
)	O
with	O
144	O
terabytes	O
of	O
memory	O
and	O
6	O
petabytes	O
of	O
storage	O
,	O
and	O
achieved	O
a	O
peak	O
performance	O
of	O
1	O
PetaFLOPS	O
.	O
</s>
<s>
JUGENE	B-Operating_System
was	O
shut	O
down	O
in	O
July	O
2012	O
and	O
replaced	O
by	O
the	O
Blue	O
Gene/Q	O
system	O
JUQUEEN	B-Device
.	O
</s>
<s>
The	O
40-rack	O
(	O
40960	O
nodes	O
,	O
163840	O
processor	O
cores	O
)	O
"	O
Intrepid	O
"	O
system	O
at	O
Argonne	O
National	O
Laboratory	O
was	O
ranked	O
#3	O
on	O
the	O
June	O
2008	O
Top	B-Operating_System
500	I-Operating_System
list	O
.	O
</s>
<s>
Lawrence	O
Livermore	O
National	O
Laboratory	O
installed	O
a	O
36-rack	O
Blue	B-Operating_System
Gene/P	I-Operating_System
installation	O
,	O
"	O
Dawn	O
"	O
,	O
in	O
2009	O
.	O
</s>
<s>
The	O
King	O
Abdullah	O
University	O
of	O
Science	O
and	O
Technology	O
(	O
KAUST	O
)	O
installed	O
a	O
16-rack	O
Blue	B-Operating_System
Gene/P	I-Operating_System
installation	O
,	O
"	O
Shaheen	B-Operating_System
"	O
,	O
in	O
2009	O
.	O
</s>
<s>
In	O
2012	O
,	O
a	O
6-rack	O
Blue	B-Operating_System
Gene/P	I-Operating_System
was	O
installed	O
at	O
Rice	O
University	O
and	O
will	O
be	O
jointly	O
administered	O
with	O
the	O
University	O
of	O
São	O
Paulo	O
.	O
</s>
<s>
A	O
2.5	O
rack	B-Application
Blue	B-Operating_System
Gene/P	I-Operating_System
system	O
is	O
the	O
central	O
processor	O
for	O
the	O
Low	O
Frequency	O
Array	O
for	O
Radio	O
astronomy	O
(	O
LOFAR	O
)	O
project	O
in	O
the	O
Netherlands	O
and	O
surrounding	O
European	O
countries	O
.	O
</s>
<s>
A	O
2-rack	O
Blue	B-Operating_System
Gene/P	I-Operating_System
was	O
installed	O
in	O
September	O
2008	O
in	O
Sofia	O
,	O
Bulgaria	O
,	O
and	O
is	O
operated	O
by	O
the	O
Bulgarian	O
Academy	O
of	O
Sciences	O
and	O
Sofia	O
University	O
.	O
</s>
<s>
In	O
2010	O
,	O
a	O
2-rack	O
(	O
8192-core	O
)	O
Blue	B-Operating_System
Gene/P	I-Operating_System
was	O
installed	O
at	O
the	O
University	O
of	O
Melbourne	O
for	O
the	O
Victorian	O
Life	O
Sciences	O
Computation	O
Initiative	O
.	O
</s>
<s>
In	O
2011	O
,	O
a	O
2-rack	O
Blue	B-Operating_System
Gene/P	I-Operating_System
was	O
installed	O
at	O
University	O
of	O
Canterbury	O
in	O
Christchurch	O
,	O
New	O
Zealand	O
.	O
</s>
<s>
In	O
2012	O
,	O
a	O
2-rack	O
Blue	B-Operating_System
Gene/P	I-Operating_System
was	O
installed	O
at	O
Rutgers	O
University	O
in	O
Piscataway	O
,	O
New	O
Jersey	O
.	O
</s>
<s>
In	O
2008	O
,	O
a	O
1-rack	O
(	O
1024	O
nodes	O
)	O
Blue	B-Operating_System
Gene/P	I-Operating_System
with	O
180	O
TB	O
of	O
storage	O
was	O
installed	O
at	O
the	O
University	O
of	O
Rochester	O
in	O
Rochester	O
,	O
New	O
York	O
.	O
</s>
<s>
The	O
first	O
Blue	B-Operating_System
Gene/P	I-Operating_System
in	O
the	O
ASEAN	O
region	O
was	O
installed	O
in	O
2010	O
at	O
the	O
Universiti	O
of	O
Brunei	O
Darussalam	O
’s	O
research	O
centre	O
,	O
the	O
UBD-IBM	O
Centre	O
.	O
</s>
<s>
In	O
2013	O
,	O
a	O
1-rack	O
Blue	B-Operating_System
Gene/P	I-Operating_System
was	O
donated	O
to	O
the	O
Department	O
of	O
Science	O
and	O
Technology	O
for	O
weather	O
forecasts	O
,	O
disaster	O
management	O
,	O
precision	O
agriculture	O
,	O
and	O
health	O
it	O
is	O
housed	O
in	O
the	O
National	O
Computer	O
Center	O
,	O
Diliman	O
,	O
Quezon	O
City	O
,	O
under	O
the	O
auspices	O
of	O
Philippine	O
Genome	O
Center	O
(	O
PGC	O
)	O
Core	O
Facility	O
for	O
Bioinformatics	O
(	O
CFB	O
)	O
at	O
UP	O
Diliman	O
,	O
Quezon	O
City	O
.	O
</s>
<s>
Veselin	O
Topalov	O
,	O
the	O
challenger	O
to	O
the	O
World	O
Chess	O
Champion	O
title	O
in	O
2010	O
,	O
confirmed	O
in	O
an	O
interview	O
that	O
he	O
had	O
used	O
a	O
Blue	B-Operating_System
Gene/P	I-Operating_System
supercomputer	B-Architecture
during	O
his	O
preparation	O
for	O
the	O
match	O
.	O
</s>
<s>
The	O
Blue	B-Operating_System
Gene/P	I-Operating_System
computer	O
has	O
been	O
used	O
to	O
simulate	O
approximately	O
one	O
percent	O
of	O
a	O
human	O
cerebral	O
cortex	O
,	O
containing	O
1.6	O
billion	O
neurons	O
with	O
approximately	O
9	O
trillion	O
connections	O
.	O
</s>
<s>
The	O
IBM	B-Device
Kittyhawk	I-Device
project	O
team	O
has	O
ported	O
Linux	B-Application
to	O
the	O
compute	O
nodes	O
and	O
demonstrated	O
generic	O
Web	O
2.0	O
workloads	O
running	O
at	O
scale	O
on	O
a	O
Blue	B-Operating_System
Gene/P	I-Operating_System
.	O
</s>
<s>
Their	O
paper	O
,	O
published	O
in	O
the	O
ACM	O
Operating	B-General_Concept
Systems	I-General_Concept
Review	O
,	O
describes	O
a	O
kernel	O
driver	O
that	O
tunnels	O
Ethernet	O
over	O
the	O
tree	O
network	O
,	O
which	O
results	O
in	O
all-to-all	O
TCP/IP	B-Protocol
connectivity	O
.	O
</s>
<s>
Running	O
standard	O
Linux	B-Application
software	O
like	O
MySQL	B-Application
,	O
their	O
performance	O
results	O
on	O
SpecJBB	O
rank	O
among	O
the	O
highest	O
on	O
record	O
.	O
</s>
<s>
In	O
2011	O
,	O
a	O
Rutgers	O
University	O
/	O
IBM	O
/	O
University	O
of	O
Texas	O
team	O
linked	O
the	O
KAUST	O
Shaheen	B-Operating_System
installation	O
together	O
with	O
a	O
Blue	B-Operating_System
Gene/P	I-Operating_System
installation	O
at	O
the	O
IBM	O
Watson	O
Research	O
Center	O
into	O
a	O
"	O
federated	O
high	B-Architecture
performance	I-Architecture
computing	I-Architecture
cloud	O
"	O
,	O
winning	O
the	O
IEEE	O
SCALE	O
2011	O
challenge	O
with	O
an	O
oil	O
reservoir	O
optimization	O
application	O
.	O
</s>
<s>
The	O
third	O
supercomputer	B-Architecture
design	O
in	O
the	O
Blue	B-Operating_System
Gene	I-Operating_System
series	O
,	O
Blue	O
Gene/Q	O
has	O
a	O
peak	O
performance	O
of	O
20	O
Petaflops	O
,	O
reaching	O
LINPACK	B-Device
benchmarks	I-Device
of	O
17	O
Petaflops	O
.	O
</s>
<s>
Blue	O
Gene/Q	O
continues	O
to	O
expand	O
and	O
enhance	O
the	O
Blue	B-Operating_System
Gene/L	I-Operating_System
and	O
/P	O
architectures	O
.	O
</s>
<s>
The	O
64-bit	B-Device
A2	B-Device
processor	O
cores	O
are	O
4-way	O
simultaneously	B-Operating_System
multithreaded	I-Operating_System
,	O
and	O
run	O
at	O
1.6GHz	O
.	O
</s>
<s>
Each	O
processor	O
core	O
has	O
a	O
SIMD	B-Device
quad-vector	O
double-precision	O
floating-point	B-General_Concept
unit	I-General_Concept
(	O
IBM	O
QPX	O
)	O
.	O
</s>
<s>
16	O
Processor	O
cores	O
are	O
used	O
for	O
computing	O
,	O
and	O
a	O
17th	O
core	O
for	O
operating	B-General_Concept
system	I-General_Concept
assist	O
functions	O
such	O
as	O
interrupts	B-Application
,	O
asynchronous	B-Architecture
I/O	I-Architecture
,	O
MPI	B-Application
pacing	O
and	O
RAS	B-General_Concept
.	O
</s>
<s>
The	O
processor	O
cores	O
are	O
linked	O
by	O
a	O
crossbar	O
switch	O
to	O
a	O
32MB	O
eDRAM	O
L2	O
cache	B-General_Concept
,	O
operating	O
at	O
half	O
core	O
speed	O
.	O
</s>
<s>
The	O
L2	O
cache	B-General_Concept
is	O
multi-versioned	O
,	O
supporting	O
transactional	B-Operating_System
memory	I-Operating_System
and	O
speculative	B-General_Concept
execution	I-General_Concept
,	O
and	O
has	O
hardware	O
support	O
for	O
atomic	B-General_Concept
operations	I-General_Concept
.	O
</s>
<s>
L2	O
cache	B-General_Concept
misses	O
are	O
handled	O
by	O
two	O
built-in	O
DDR3	O
memory	O
controllers	O
running	O
at	O
1.33GHz	O
.	O
</s>
<s>
Separate	O
I/O	O
drawers	O
,	O
placed	O
at	O
the	O
top	O
of	O
a	O
rack	B-Application
or	O
in	O
a	O
separate	O
rack	B-Application
,	O
are	O
air	O
cooled	O
and	O
contain	O
8	O
compute	O
cards	O
and	O
8	O
PCIe	O
expansion	O
slots	O
for	O
InfiniBand	B-Architecture
or	O
10	O
Gigabit	O
Ethernet	O
networking	O
.	O
</s>
<s>
At	O
the	O
time	O
of	O
the	O
Blue	O
Gene/Q	O
system	O
announcement	O
in	O
November	O
2011	O
,	O
an	O
initial	O
4-rack	O
Blue	O
Gene/Q	O
system	O
(	O
4096	O
nodes	O
,	O
65536	O
user	O
processor	O
cores	O
)	O
achieved	O
#17	O
in	O
the	O
TOP500	B-Operating_System
list	O
with	O
677.1	O
TeraFLOPS	O
Linpack	B-Application
,	O
outperforming	O
the	O
original	O
2007	O
104-rack	O
BlueGene/L	B-Operating_System
installation	O
described	O
above	O
.	O
</s>
<s>
The	O
same	O
4-rack	O
system	O
achieved	O
the	O
top	O
position	O
in	O
the	O
Graph500	B-Device
list	O
with	O
over	O
250	O
GTEPS	O
(	O
giga	O
traversed	B-Operating_System
edges	I-Operating_System
per	I-Operating_System
second	I-Operating_System
)	O
.	O
</s>
<s>
Blue	O
Gene/Q	O
systems	O
also	O
topped	O
the	O
Green500	B-Application
list	O
of	O
most	O
energy	O
efficient	O
supercomputers	B-Architecture
with	O
up	O
to	O
2.1GFLOPS/W	O
.	O
</s>
<s>
In	O
June	O
2012	O
,	O
Blue	O
Gene/Q	O
installations	O
took	O
the	O
top	O
positions	O
in	O
all	O
three	O
lists	O
:	O
TOP500	B-Operating_System
,	O
Graph500	B-Device
and	O
Green500	B-Application
.	O
</s>
<s>
Per	O
June	O
2012	O
,	O
the	O
TOP500	B-Operating_System
list	O
contained	O
20	O
Blue	O
Gene/Q	O
installations	O
of	O
1/2	O
-rack	O
(	O
512	O
nodes	O
,	O
8192	O
processor	O
cores	O
,	O
86.35TFLOPS	O
Linpack	B-Application
)	O
and	O
larger	O
.	O
</s>
<s>
At	O
a	O
(	O
size-independent	O
)	O
power	O
efficiency	O
of	O
about	O
2.1	O
GFLOPS/W	O
,	O
all	O
these	O
systems	O
also	O
populated	O
the	O
top	O
of	O
the	O
June	O
2012	O
Green	B-Application
500	I-Application
list	O
.	O
</s>
<s>
A	O
Blue	O
Gene/Q	O
system	O
called	O
Sequoia	B-Device
was	O
delivered	O
to	O
the	O
Lawrence	O
Livermore	O
National	O
Laboratory	O
(	O
LLNL	O
)	O
beginning	O
in	O
2011	O
and	O
was	O
fully	O
deployed	O
in	O
June	O
2012	O
.	O
</s>
<s>
It	O
is	O
part	O
of	O
the	O
Advanced	B-General_Concept
Simulation	I-General_Concept
and	I-General_Concept
Computing	I-General_Concept
Program	I-General_Concept
running	O
nuclear	O
simulations	O
and	O
advanced	O
scientific	O
research	O
.	O
</s>
<s>
In	O
June	O
2012	O
,	O
the	O
system	O
was	O
ranked	O
as	O
the	O
world	O
's	O
fastest	B-Operating_System
supercomputer	I-Operating_System
.	O
</s>
<s>
at	O
20.1PFLOPS	O
peak	O
,	O
16.32PFLOPS	O
sustained	O
(	O
Linpack	B-Application
)	O
,	O
drawing	O
up	O
to	O
7.9	O
megawatts	O
of	O
power	O
.	O
</s>
<s>
In	O
June	O
2013	O
,	O
its	O
performance	O
is	O
listed	O
at	O
17.17PFLOPS	O
sustained	O
(	O
Linpack	B-Application
)	O
.	O
</s>
<s>
A	O
10	O
PFLOPS	O
(	O
peak	O
)	O
Blue	O
Gene/Q	O
system	O
called	O
Mira	B-Device
was	O
installed	O
at	O
Argonne	O
National	O
Laboratory	O
in	O
the	O
in	O
2012	O
.	O
</s>
<s>
JUQUEEN	B-Device
at	O
the	O
Forschungzentrum	O
Jülich	O
is	O
a	O
28-rack	O
Blue	O
Gene/Q	O
system	O
,	O
and	O
was	O
from	O
June	O
2013	O
to	O
November	O
2015	O
the	O
highest	O
ranked	O
machine	O
in	O
Europe	O
in	O
the	O
Top500	B-Operating_System
.	O
</s>
<s>
Vulcan	O
at	O
Lawrence	O
Livermore	O
National	O
Laboratory	O
(	O
LLNL	O
)	O
is	O
a	O
24-rack	O
,	O
5PFLOPS	O
(	O
peak	O
)	O
,	O
Blue	O
Gene/Q	O
system	O
that	O
was	O
commissioned	O
in	O
2012	O
and	O
decommissioned	O
in	O
2019	O
.	O
</s>
<s>
Vulcan	O
served	O
Lab-industry	O
projects	O
through	O
Livermore	O
's	O
High	B-Architecture
Performance	I-Architecture
Computing	I-Architecture
(	O
HPC	O
)	O
Innovation	O
Center	O
as	O
well	O
as	O
academic	O
collaborations	O
in	O
support	O
of	O
DOE/National	O
Nuclear	O
Security	O
Administration	O
(	O
NNSA	O
)	O
missions	O
.	O
</s>
<s>
Fermi	O
at	O
the	O
CINECA	B-Device
Supercomputing	B-Architecture
facility	O
,	O
Bologna	O
,	O
Italy	O
,	O
is	O
a	O
10-rack	O
,	O
2PFLOPS	O
(	O
peak	O
)	O
,	O
Blue	O
Gene/Q	O
system	O
.	O
</s>
<s>
A	O
five	O
rack	B-Application
Blue	O
Gene/Q	O
system	O
with	O
additional	O
compute	O
hardware	O
called	O
AMOS	O
was	O
installed	O
at	O
Rensselaer	O
Polytechnic	O
Institute	O
in	O
2013	O
.	O
</s>
<s>
The	O
system	O
was	O
rated	O
at	O
1048.6	O
teraflops	O
,	O
the	O
most	O
powerful	O
supercomputer	B-Architecture
at	O
any	O
private	O
university	O
,	O
and	O
third	O
most	O
powerful	O
supercomputer	B-Architecture
among	O
all	O
universities	O
in	O
2014	O
.	O
</s>
<s>
This	O
system	O
is	O
part	O
of	O
the	O
,	O
which	O
is	O
dedicated	O
to	O
the	O
application	O
of	O
high-performance	B-Architecture
computing	I-Architecture
to	O
research	O
programs	O
in	O
the	O
health	O
sciences	O
.	O
</s>
<s>
The	O
system	O
consists	O
of	O
a	O
single	O
rack	B-Application
(	O
1,024	O
compute	O
nodes	O
)	O
with	O
400	O
TB	O
of	O
high-performance	O
storage	O
.	O
</s>
<s>
A	O
209	O
TFLOPS	O
peak	O
(	O
172TFLOPS	O
LINPACK	B-Application
)	O
Blue	O
Gene/Q	O
system	O
called	O
Lemanicus	O
was	O
installed	O
at	O
the	O
EPFL	O
in	O
March	O
2013	O
.	O
</s>
<s>
The	O
system	O
consists	O
of	O
a	O
single	O
rack	B-Application
(	O
1,024	O
compute	O
nodes	O
)	O
with	O
2.1PB	O
of	O
IBM	O
GPFS-GSS	O
storage	O
.	O
</s>
<s>
A	O
half-rack	O
Blue	O
Gene/Q	O
system	O
,	O
with	O
about	O
100TFLOPS	O
(	O
peak	O
)	O
,	O
called	O
Cumulus	O
was	O
installed	O
at	O
A*STAR	O
Computational	O
Resource	O
Centre	O
,	O
Singapore	O
,	O
at	O
early	O
2011	O
.	O
</s>
<s>
The	O
cosmology	O
simulation	O
framework	O
HACC	O
achieved	O
almost	O
14	O
petaflops	O
with	O
a	O
3.6	O
trillion	O
particle	O
benchmark	O
run	O
,	O
while	O
the	O
Cardioid	O
code	O
,	O
which	O
models	O
the	O
electrophysiology	O
of	O
the	O
human	O
heart	O
,	O
achieved	O
nearly	O
12	O
petaflops	O
with	O
a	O
near	O
real-time	O
simulation	O
,	O
both	O
on	O
Sequoia	B-Device
.	O
</s>
<s>
A	O
fully	O
compressible	O
flow	O
solver	O
has	O
also	O
achieved	O
14.4PFLOP/s	O
(	O
originally	O
11PFLOP/s	O
)	O
on	O
Sequoia	B-Device
,	O
72%	O
of	O
the	O
machine	O
's	O
nominal	O
peak	O
performance	O
.	O
</s>
