<s>
Anton	B-Device
is	O
a	O
massively	O
parallel	O
supercomputer	O
designed	O
and	O
built	O
by	O
D	B-Application
.	I-Application
E	I-Application
.	I-Application
Shaw	I-Application
Research	I-Application
in	O
New	O
York	O
,	O
first	O
running	O
in	O
2008	O
.	O
</s>
<s>
An	O
Anton	B-Device
machine	O
consists	O
of	O
a	O
substantial	O
number	O
of	O
application-specific	O
integrated	O
circuits	O
(	O
ASICs	O
)	O
,	O
interconnected	O
by	O
a	O
specialized	O
high-speed	O
,	O
three-dimensional	O
torus	B-Operating_System
network	I-Operating_System
.	O
</s>
<s>
Unlike	O
earlier	O
special-purpose	O
systems	O
for	O
MD	O
simulations	O
,	O
such	O
as	O
MDGRAPE-3	B-Algorithm
developed	O
by	O
RIKEN	O
in	O
Japan	O
,	O
Anton	B-Device
runs	O
its	O
computations	O
entirely	O
on	O
specialized	O
ASICs	O
,	O
instead	O
of	O
dividing	O
the	O
computation	O
between	O
specialized	O
ASICs	O
and	O
general-purpose	O
host	O
processors	O
.	O
</s>
<s>
Each	O
Anton	B-Device
ASIC	O
contains	O
two	O
computational	O
subsystems	O
.	O
</s>
<s>
This	O
subsystem	O
contains	O
32	O
deeply	O
pipelined	O
modules	O
running	O
at	O
800MHz	O
arranged	O
much	O
like	O
a	O
systolic	B-Architecture
array	I-Architecture
.	O
</s>
<s>
This	O
subsystem	O
contains	O
four	O
general-purpose	O
Tensilica	B-General_Concept
cores	O
(	O
each	O
with	O
cache	O
and	O
scratchpad	O
memory	O
)	O
and	O
eight	O
specialized	O
but	O
programmable	O
SIMD	B-Device
cores	O
called	O
geometry	O
cores	O
.	O
</s>
<s>
Anton	B-Device
's	O
network	O
is	O
a	O
3D	O
torus	O
and	O
thus	O
each	O
chip	O
has	O
6	O
inter-node	O
links	O
with	O
a	O
total	O
in+out	O
bandwidth	O
of	O
607.2	O
Gbit/s	O
.	O
</s>
<s>
The	O
per-hop	O
latency	O
in	O
Anton	B-Device
's	O
network	O
is	O
50ns	O
.	O
</s>
<s>
The	O
performance	O
of	O
a	O
512-node	O
Anton	B-Device
machine	O
is	O
over	O
17,000	O
nanoseconds	O
of	O
simulated	O
time	O
per	O
day	O
for	O
a	O
protein-water	O
system	O
consisting	O
of	O
23,558	O
atoms	O
.	O
</s>
<s>
The	O
first	O
512-node	O
Anton	B-Device
machine	O
became	O
operational	O
in	O
October	O
2008	O
.	O
</s>
<s>
The	O
multiple	O
petaFLOP	O
,	O
distributed-computing	O
project	O
Folding	B-Application
@home	I-Application
has	O
achieved	O
similar	O
aggregate	O
ensemble	O
simulation	O
timescales	O
,	O
comparable	O
to	O
the	O
total	O
time	O
of	O
a	O
single	O
continuous	O
simulation	O
on	O
Anton	B-Device
,	O
specifically	O
achieving	O
the	O
1.5-millisecond	O
range	O
in	O
January	O
2010	O
.	O
</s>
<s>
The	O
Anton	B-Device
supercomputer	O
is	O
named	O
after	O
Anton	B-Device
van	O
Leeuwenhoek	O
,	O
who	O
is	O
often	O
referred	O
to	O
as	O
"	O
the	O
father	O
of	O
microscopy	O
"	O
because	O
he	O
built	O
high-precision	O
optical	O
instruments	O
and	O
used	O
them	O
to	O
visualize	O
a	O
wide	O
variety	O
of	O
organisms	O
and	O
cell	O
types	O
for	O
the	O
first	O
time	O
.	O
</s>
<s>
The	O
ANTON	B-Device
2	O
machine	O
with	O
four	O
512	O
nodes	O
and	O
its	O
substantially	O
increased	O
speed	O
and	O
problem	O
size	O
has	O
been	O
described	O
.	O
</s>
<s>
The	O
National	O
Institutes	O
of	O
Health	O
have	O
supported	O
an	O
ANTON	B-Device
for	O
the	O
biomedical	O
research	O
community	O
at	O
the	O
Pittsburgh	O
Supercomputing	O
Center	O
,	O
Carnegie	O
Mellon	O
University	O
,	O
and	O
currently	O
(	O
8/20	O
)	O
continues	O
with	O
an	O
ANTON	B-Device
2	O
system	O
.	O
</s>
