<s>
A	O
graphics	B-Architecture
processing	I-Architecture
unit	I-Architecture
(	O
GPU	B-Architecture
)	O
is	O
a	O
specialized	O
electronic	O
circuit	O
designed	O
to	O
manipulate	O
and	O
alter	O
memory	B-General_Concept
to	O
accelerate	O
the	O
creation	O
of	O
images	B-Algorithm
in	O
a	O
frame	B-Algorithm
buffer	I-Algorithm
intended	O
for	O
output	O
to	O
a	O
display	B-Device
device	I-Device
.	O
</s>
<s>
GPUs	B-Architecture
are	O
used	O
in	O
embedded	B-Architecture
systems	I-Architecture
,	O
mobile	O
phones	O
,	O
personal	B-Device
computers	I-Device
,	O
workstations	B-Device
,	O
and	O
game	O
consoles	O
.	O
</s>
<s>
Modern	O
GPUs	B-Architecture
are	O
efficient	O
at	O
manipulating	O
computer	O
graphics	O
and	O
image	B-Algorithm
processing	I-Algorithm
.	O
</s>
<s>
Their	O
parallel	B-Operating_System
structure	I-Operating_System
makes	O
them	O
more	O
efficient	O
than	O
general-purpose	O
central	B-General_Concept
processing	I-General_Concept
units	I-General_Concept
(	O
CPUs	B-General_Concept
)	O
for	O
algorithms	O
that	O
process	O
large	O
blocks	O
of	O
data	O
in	O
parallel	B-Operating_System
.	O
</s>
<s>
In	O
a	O
personal	B-Device
computer	I-Device
,	O
a	O
GPU	B-Architecture
can	O
be	O
present	O
on	O
a	O
video	B-Device
card	I-Device
or	O
embedded	O
on	O
the	O
motherboard	B-Device
.	O
</s>
<s>
In	O
some	O
CPUs	B-General_Concept
,	O
they	O
are	O
embedded	O
on	O
the	O
CPU	O
die	O
.	O
</s>
<s>
Arcade	O
system	B-Device
boards	I-Device
have	O
used	O
specialized	O
graphics	O
circuits	O
since	O
the	O
1970s	O
.	O
</s>
<s>
In	O
early	O
video	O
game	O
hardware	B-Architecture
,	O
RAM	B-Architecture
for	O
frame	B-Algorithm
buffers	I-Algorithm
was	O
expensive	O
,	O
so	O
video	B-General_Concept
chips	I-General_Concept
composited	O
data	O
together	O
as	O
the	O
display	O
was	O
being	O
scanned	O
out	O
on	O
the	O
monitor	O
.	O
</s>
<s>
A	O
specialized	O
barrel	O
shifter	O
circuit	O
helped	O
the	O
CPU	O
animate	O
the	O
framebuffer	B-Algorithm
graphics	O
for	O
various	O
1970s	O
arcade	B-Application
video	I-Application
games	I-Application
from	O
Midway	O
and	O
Taito	O
,	O
such	O
as	O
Gun	B-Application
Fight	I-Application
(	O
1975	O
)	O
,	O
Sea	B-Application
Wolf	I-Application
(	O
1976	O
)	O
,	O
and	O
Space	B-Application
Invaders	I-Application
(	O
1978	O
)	O
.	O
</s>
<s>
The	O
Namco	O
Galaxian	O
arcade	O
system	O
in	O
1979	O
used	O
specialized	O
graphics	B-General_Concept
hardware	I-General_Concept
that	O
supported	O
RGB	O
color	O
,	O
multi-colored	O
sprites	O
,	O
and	O
tilemap	B-Application
backgrounds	O
.	O
</s>
<s>
The	O
Galaxian	O
hardware	B-Architecture
was	O
widely	O
used	O
during	O
the	O
golden	O
age	O
of	O
arcade	B-Application
video	I-Application
games	I-Application
,	O
by	O
game	O
companies	O
such	O
as	O
Namco	O
,	O
Centuri	O
,	O
Gremlin	O
,	O
Irem	O
,	O
Konami	O
,	O
Midway	O
,	O
Nichibutsu	O
,	O
Sega	O
,	O
and	O
Taito	O
.	O
</s>
<s>
The	O
Atari	B-General_Concept
2600	I-General_Concept
in	O
1977	O
used	O
a	O
video	O
shifter	O
called	O
the	O
Television	B-General_Concept
Interface	I-General_Concept
Adaptor	I-General_Concept
.	O
</s>
<s>
Atari	B-Device
8-bit	I-Device
computers	I-Device
(	O
1979	O
)	O
had	O
ANTIC	B-General_Concept
,	O
a	O
video	O
processor	O
which	O
interpreted	O
instructions	O
describing	O
a	O
"	O
display	O
list	O
"	O
—	O
the	O
way	O
the	O
scan	B-Algorithm
lines	I-Algorithm
map	O
to	O
specific	O
bitmapped	B-Algorithm
or	O
character	O
modes	O
and	O
where	O
the	O
memory	B-General_Concept
is	O
stored	O
(	O
so	O
there	O
did	O
not	O
need	O
to	O
be	O
a	O
contiguous	O
frame	B-Algorithm
buffer	I-Algorithm
)	O
.	O
</s>
<s>
6502	B-General_Concept
machine	B-Language
code	I-Language
subroutines	O
could	O
be	O
triggered	O
on	O
scan	B-Algorithm
lines	I-Algorithm
by	O
setting	O
a	O
bit	O
on	O
a	O
display	O
list	O
instruction	O
.	O
</s>
<s>
ANTIC	B-General_Concept
also	O
supported	O
smooth	O
vertical	O
and	O
horizontal	O
scrolling	O
independent	O
of	O
the	O
CPU	O
.	O
</s>
<s>
The	O
NEC	B-Device
µPD7220	I-Device
was	O
the	O
first	O
implementation	O
of	O
a	O
PC	O
graphics	O
display	O
processor	O
as	O
a	O
single	O
large-scale	O
integration	O
(	O
LSI	O
)	O
integrated	O
circuit	O
chip	O
.	O
</s>
<s>
This	O
enabled	O
the	O
design	O
of	O
low-cost	O
,	O
high-performance	O
video	O
graphics	B-Device
cards	I-Device
such	O
as	O
those	O
from	O
Number	O
Nine	O
Visual	O
Technology	O
.	O
</s>
<s>
It	O
became	O
the	O
best-known	O
GPU	B-Architecture
until	O
the	O
mid-1980s	O
.	O
</s>
<s>
It	O
was	O
the	O
first	O
fully	O
integrated	O
VLSI	O
(	O
very	O
large-scale	O
integration	O
)	O
metal	B-Architecture
–	I-Architecture
oxide	I-Architecture
–	I-Architecture
semiconductor	I-Architecture
(	O
NMOS	B-Algorithm
)	O
graphics	O
display	O
processor	O
for	O
PCs	B-Device
,	O
supported	O
up	O
to	O
1024×1024	B-Device
resolution	I-Device
,	O
and	O
laid	O
the	O
foundations	O
for	O
the	O
emerging	O
PC	O
graphics	O
market	O
.	O
</s>
<s>
It	O
was	O
used	O
in	O
a	O
number	O
of	O
graphics	B-Device
cards	I-Device
and	O
was	O
licensed	O
for	O
clones	O
such	O
as	O
the	O
Intel	O
82720	O
,	O
the	O
first	O
of	O
Intel	B-Device
's	I-Device
graphics	I-Device
processing	I-Device
units	I-Device
.	O
</s>
<s>
The	O
Williams	O
Electronics	O
arcade	O
games	O
Robotron	B-Application
2084	I-Application
,	O
Joust	B-Application
,	O
Sinistar	B-Application
,	O
and	O
Bubbles	B-Application
,	O
all	O
released	O
in	O
1982	O
,	O
contain	O
custom	O
blitter	B-General_Concept
chips	I-General_Concept
for	O
operating	O
on	O
16-color	O
bitmaps	O
.	O
</s>
<s>
In	O
1984	O
,	O
Hitachi	O
released	O
ARTC	O
HD63484	O
,	O
the	O
first	O
major	O
CMOS	B-Device
graphics	B-Architecture
processor	I-Architecture
for	O
PC	O
.	O
</s>
<s>
The	O
ARTC	O
could	O
display	O
up	O
to	O
4K	B-Architecture
resolution	I-Architecture
when	O
in	O
monochrome	O
mode	O
.	O
</s>
<s>
It	O
was	O
used	O
in	O
a	O
number	O
of	O
graphics	B-Device
cards	I-Device
and	O
terminals	O
during	O
the	O
late	O
1980s	O
.	O
</s>
<s>
In	O
1985	O
,	O
the	O
Amiga	B-Device
was	O
released	O
with	O
a	O
custom	O
graphics	B-General_Concept
chip	I-General_Concept
including	O
a	O
blitter	B-General_Concept
for	O
bitmap	O
manipulation	O
,	O
line	O
drawing	O
,	O
and	O
area	O
fill	O
.	O
</s>
<s>
It	O
also	O
included	O
a	O
coprocessor	B-General_Concept
with	O
its	O
own	O
simple	O
instruction	O
set	O
,	O
that	O
was	O
capable	O
of	O
manipulating	O
graphics	B-General_Concept
hardware	I-General_Concept
registers	O
in	O
sync	O
with	O
the	O
video	O
beam	O
(	O
e.g.	O
</s>
<s>
for	O
per-scanline	O
palette	O
switches	O
,	O
sprite	O
multiplexing	O
,	O
and	O
hardware	B-Architecture
windowing	O
)	O
,	O
or	O
driving	O
the	O
blitter	B-General_Concept
.	O
</s>
<s>
In	O
1986	O
,	O
Texas	O
Instruments	O
released	O
the	O
TMS34010	B-General_Concept
,	O
the	O
first	O
fully	O
programmable	O
graphics	B-Architecture
processor	I-Architecture
.	O
</s>
<s>
During	O
1990	O
–	O
92	O
,	O
this	O
chip	O
became	O
the	O
basis	O
of	O
the	O
Texas	B-Device
Instruments	I-Device
Graphics	I-Device
Architecture	I-Device
(	O
"	O
TIGA	O
"	O
)	O
Windows	B-Device
accelerator	I-Device
cards	O
.	O
</s>
<s>
In	O
1987	O
,	O
the	O
IBM	B-Device
8514	I-Device
graphics	O
system	O
was	O
released	O
.	O
</s>
<s>
It	O
was	O
one	O
of	O
the	O
first	O
video	B-Device
cards	I-Device
for	O
IBM	O
PC	O
compatibles	O
to	O
implement	O
fixed-function	O
2D	O
primitives	O
in	O
electronic	O
hardware	B-Architecture
.	O
</s>
<s>
Sharp	O
's	O
X68000	B-Device
,	O
released	O
in	O
1987	O
,	O
used	O
a	O
custom	O
graphics	O
chipset	O
with	O
a	O
65,536	O
color	O
palette	O
and	O
hardware	B-Architecture
support	O
for	O
sprites	O
,	O
scrolling	O
,	O
and	O
multiple	O
playfields	O
.	O
</s>
<s>
It	O
served	O
as	O
a	O
development	O
machine	O
for	O
Capcom	O
's	O
CP	B-Application
System	I-Application
arcade	O
board	O
.	O
</s>
<s>
Fujitsu	O
's	O
FM	B-Device
Towns	I-Device
computer	O
,	O
released	O
in	O
1989	O
,	O
had	O
support	O
for	O
a	O
16,777,216	O
color	O
palette	O
.	O
</s>
<s>
In	O
1988	O
,	O
the	O
first	O
dedicated	O
polygonal	O
3D	O
graphics	B-Device
boards	I-Device
were	O
introduced	O
in	O
arcades	O
with	O
the	O
Namco	B-Application
System	I-Application
21	I-Application
and	O
Taito	O
Air	O
System	O
.	O
</s>
<s>
IBM	O
introduced	O
its	O
proprietary	B-Application
Video	B-Protocol
Graphics	I-Protocol
Array	I-Protocol
(	O
VGA	B-Protocol
)	O
display	O
standard	O
in	O
1987	O
,	O
with	O
a	O
maximum	O
resolution	O
of	O
640×480	O
pixels	B-Algorithm
.	O
</s>
<s>
In	O
November	O
1988	O
,	O
NEC	O
Home	B-Device
Electronics	O
announced	O
its	O
creation	O
of	O
the	O
Video	O
Electronics	O
Standards	O
Association	O
(	O
VESA	O
)	O
to	O
develop	O
and	O
promote	O
a	O
Super	O
VGA	B-Protocol
(	O
SVGA	O
)	O
computer	B-Device
display	I-Device
standard	O
as	O
a	O
successor	O
to	O
VGA	B-Protocol
.	O
</s>
<s>
Super	O
VGA	B-Protocol
enabled	O
graphics	O
display	O
resolutions	O
up	O
to	O
800×600	O
pixels	B-Algorithm
,	O
a	O
36%	O
increase	O
.	O
</s>
<s>
The	O
86C911	O
spawned	O
a	O
host	O
of	O
imitators	O
:	O
by	O
1995	O
,	O
all	O
major	O
PC	O
graphics	B-General_Concept
chip	I-General_Concept
makers	O
had	O
added	O
2D	O
acceleration	O
support	O
to	O
their	O
chips	O
.	O
</s>
<s>
Fixed-function	O
Windows	B-Device
accelerators	I-Device
surpassed	O
expensive	O
general-purpose	O
graphics	B-Architecture
coprocessors	I-Architecture
in	O
Windows	B-Application
performance	O
,	O
and	O
such	O
coprocessors	B-General_Concept
faded	O
away	O
from	O
the	O
PC	O
market	O
.	O
</s>
<s>
Throughout	O
the	O
1990s	O
,	O
2D	O
GUI	B-Application
acceleration	O
evolved	O
.	O
</s>
<s>
As	O
manufacturing	O
capabilities	O
improved	O
,	O
so	O
did	O
the	O
level	O
of	O
integration	O
of	O
graphics	B-General_Concept
chips	I-General_Concept
.	O
</s>
<s>
Additional	O
application	B-Application
programming	I-Application
interfaces	I-Application
(	O
APIs	B-General_Concept
)	O
arrived	O
for	O
a	O
variety	O
of	O
tasks	O
,	O
such	O
as	O
Microsoft	O
's	O
WinG	O
graphics	B-Library
library	I-Library
for	O
Windows	B-Application
3.x	I-Application
,	O
and	O
their	O
later	O
DirectDraw	B-Device
interface	O
for	O
hardware	B-General_Concept
acceleration	I-General_Concept
of	O
2D	O
games	O
in	O
Windows	B-Application
95	I-Application
and	O
later	O
.	O
</s>
<s>
In	O
the	O
early	O
-	O
and	O
mid-1990s	O
,	O
real-time	B-General_Concept
3D	O
graphics	O
became	O
increasingly	O
common	O
in	O
arcade	O
,	O
computer	O
,	O
and	O
console	O
games	O
,	O
which	O
led	O
to	O
increasing	O
public	O
demand	O
for	O
hardware-accelerated	B-General_Concept
3D	O
graphics	O
.	O
</s>
<s>
Early	O
examples	O
of	O
mass-market	O
3D	O
graphics	B-General_Concept
hardware	I-General_Concept
can	O
be	O
found	O
in	O
arcade	O
system	B-Device
boards	I-Device
such	O
as	O
the	O
Sega	O
Model	O
1	O
,	O
Namco	B-Application
System	I-Application
22	I-Application
,	O
and	O
Sega	O
Model	O
2	O
,	O
and	O
the	O
fifth-generation	B-Application
video	I-Application
game	I-Application
consoles	I-Application
such	O
as	O
the	O
Saturn	B-Device
,	O
PlayStation	B-Device
,	O
and	O
Nintendo	B-Operating_System
64	I-Operating_System
.	O
</s>
<s>
Arcade	O
systems	O
such	O
as	O
the	O
Sega	O
Model	O
2	O
and	O
SGI	O
Onyx-based	O
Namco	O
Magic	O
Edge	O
Hornet	O
Simulator	O
in	O
1993	O
were	O
capable	O
of	O
hardware	B-Algorithm
T&L	I-Algorithm
(	O
transform	B-Algorithm
,	I-Algorithm
clipping	I-Algorithm
,	I-Algorithm
and	I-Algorithm
lighting	I-Algorithm
)	O
years	O
before	O
appearing	O
in	O
consumer	O
graphics	B-Device
cards	I-Device
.	O
</s>
<s>
Some	O
systems	O
used	O
DSPs	B-Architecture
to	O
accelerate	O
transformations	O
.	O
</s>
<s>
Fujitsu	O
,	O
which	O
worked	O
on	O
the	O
Sega	O
Model	O
2	O
arcade	O
system	O
,	O
began	O
working	O
on	O
integrating	O
T&L	B-Algorithm
into	O
a	O
single	O
LSI	O
solution	O
for	O
use	O
in	O
home	B-Device
computers	O
in	O
1995	O
;	O
the	O
Fujitsu	O
Pinolite	O
,	O
the	O
first	O
3D	O
geometry	O
processor	O
for	O
personal	B-Device
computers	I-Device
,	O
released	O
in	O
1997	O
.	O
</s>
<s>
The	O
first	O
hardware	B-Algorithm
T&L	I-Algorithm
GPU	B-Architecture
on	O
home	B-Device
video	I-Device
game	I-Device
consoles	I-Device
was	O
the	O
Nintendo	B-Operating_System
64	I-Operating_System
's	O
Reality	O
Coprocessor	B-General_Concept
,	O
released	O
in	O
1996	O
.	O
</s>
<s>
In	O
1997	O
,	O
Mitsubishi	O
released	O
the	O
3Dpro/2MP	B-Application
,	O
a	O
GPU	B-Architecture
capable	O
of	O
transformation	O
and	O
lighting	O
,	O
for	O
workstations	B-Device
and	O
Windows	B-Device
NT	I-Device
desktops	O
;	O
ATi	O
utilized	O
it	O
for	O
its	O
FireGL	B-Application
4000	I-Application
graphics	B-Device
card	I-Device
,	O
released	O
in	O
1997	O
.	O
</s>
<s>
The	O
term	O
"	O
GPU	B-Architecture
"	O
was	O
coined	O
by	O
Sony	O
in	O
reference	O
to	O
the	O
32-bit	O
Sony	B-Device
GPU	I-Device
(	O
designed	O
by	O
Toshiba	O
)	O
in	O
the	O
PlayStation	B-Device
video	B-Device
game	I-Device
console	I-Device
,	O
released	O
in	O
1994	O
.	O
</s>
<s>
In	O
the	O
PC	O
world	O
,	O
notable	O
failed	O
attempts	O
for	O
low-cost	O
3D	O
graphics	B-General_Concept
chips	I-General_Concept
included	O
the	O
S3	B-Device
ViRGE	I-Device
,	O
ATI	B-Device
Rage	I-Device
,	O
and	O
Matrox	B-Application
Mystique	O
.	O
</s>
<s>
These	O
chips	O
were	O
essentially	O
previous-generation	O
2D	B-Architecture
accelerators	I-Architecture
with	O
3D	O
features	O
bolted	O
on	O
.	O
</s>
<s>
Initially	O
,	O
3D	O
graphics	O
were	O
possible	O
only	O
with	O
discrete	O
boards	O
dedicated	O
to	O
accelerating	O
3D	O
functions	O
(	O
and	O
lacking	O
2D	O
GUI	B-Application
acceleration	O
entirely	O
)	O
such	O
as	O
the	O
PowerVR	B-Device
and	O
the	O
3dfx	O
Voodoo	O
.	O
</s>
<s>
However	O
,	O
as	O
manufacturing	O
technology	O
continued	O
to	O
progress	O
,	O
video	O
,	O
2D	O
GUI	B-Application
acceleration	O
,	O
and	O
3D	O
functionality	O
were	O
all	O
integrated	O
into	O
one	O
chip	O
.	O
</s>
<s>
In	O
1997	O
,	O
Rendition	O
collaborated	O
with	O
Hercules	O
and	O
Fujitsu	O
on	O
a	O
"	O
Thriller	O
Conspiracy	O
"	O
project	O
which	O
combined	O
a	O
Fujitsu	O
FXG-1	O
Pinolite	O
geometry	O
processor	O
with	O
a	O
Vérité	O
V2200	O
core	O
to	O
create	O
a	O
graphics	B-Device
card	I-Device
with	O
a	O
full	O
T&L	B-Algorithm
engine	O
years	O
before	O
Nvidia	O
's	O
GeForce	B-Device
256	I-Device
.	O
</s>
<s>
OpenGL	B-Application
appeared	O
in	O
the	O
early	O
'	O
90s	O
as	O
a	O
professional	O
graphics	B-Library
API	I-Library
,	O
but	O
originally	O
suffered	O
from	O
performance	O
issues	O
which	O
allowed	O
the	O
Glide	B-Operating_System
API	I-Operating_System
to	O
become	O
a	O
dominant	O
force	O
on	O
the	O
PC	O
in	O
the	O
late	O
'	O
90s	O
.	O
</s>
<s>
These	O
issues	O
were	O
quickly	O
overcome	O
and	O
the	O
Glide	B-Operating_System
API	I-Operating_System
fell	O
by	O
the	O
wayside	O
.	O
</s>
<s>
Software	O
implementations	O
of	O
OpenGL	B-Application
were	O
common	O
during	O
this	O
time	O
,	O
although	O
the	O
influence	O
of	O
OpenGL	B-Application
eventually	O
led	O
to	O
widespread	O
hardware	B-Architecture
support	O
.	O
</s>
<s>
DirectX	B-Application
became	O
popular	O
among	O
Windows	B-Application
game	O
developers	O
during	O
the	O
late	O
90s	O
.	O
</s>
<s>
Unlike	O
OpenGL	B-Application
,	O
Microsoft	O
insisted	O
on	O
providing	O
strict	O
one-to-one	O
support	O
of	O
hardware	B-Architecture
.	O
</s>
<s>
The	O
approach	O
made	O
DirectX	B-Application
less	O
popular	O
as	O
a	O
standalone	O
graphics	B-Library
API	I-Library
initially	O
,	O
since	O
many	O
GPUs	B-Architecture
provided	O
their	O
own	O
specific	O
features	O
,	O
which	O
existing	O
OpenGL	B-Application
applications	O
were	O
already	O
able	O
to	O
benefit	O
from	O
,	O
leaving	O
DirectX	B-Application
often	O
one	O
generation	O
behind	O
.	O
</s>
<s>
(	O
See	O
:	O
Comparison	B-Application
of	I-Application
OpenGL	I-Application
and	I-Application
Direct3D	I-Application
.	O
)	O
</s>
<s>
Microsoft	O
began	O
to	O
work	O
more	O
closely	O
with	O
hardware	B-Architecture
developers	O
and	O
started	O
to	O
target	O
the	O
releases	O
of	O
DirectX	B-Application
to	O
coincide	O
with	O
those	O
of	O
the	O
supporting	O
graphics	B-General_Concept
hardware	I-General_Concept
.	O
</s>
<s>
Direct3D	B-Application
5.0	O
was	O
the	O
first	O
version	O
of	O
the	O
API	B-General_Concept
to	O
gain	O
widespread	O
adoption	O
in	O
the	O
gaming	O
market	O
,	O
and	O
it	O
competed	O
directly	O
with	O
more-hardware-specific	O
,	O
often	O
proprietary	B-Application
,	O
graphics	B-Library
libraries	I-Library
,	O
while	O
OpenGL	B-Application
maintained	O
a	O
strong	O
following	O
.	O
</s>
<s>
Direct3D	B-Application
7.0	O
introduced	O
support	O
for	O
hardware-accelerated	B-General_Concept
transform	B-Algorithm
and	I-Algorithm
lighting	I-Algorithm
(	O
T&L	B-Algorithm
)	O
for	O
Direct3D	B-Application
;	O
OpenGL	B-Application
had	O
this	O
capability	O
from	O
its	O
inception	O
.	O
</s>
<s>
3D	B-Architecture
accelerator	I-Architecture
cards	I-Architecture
moved	O
beyond	O
being	O
simple	O
rasterizers	O
to	O
add	O
another	O
significant	O
hardware	B-Architecture
stage	O
to	O
the	O
3D	O
rendering	O
pipeline	O
.	O
</s>
<s>
The	O
Nvidia	B-Application
GeForce	I-Application
256	O
(	O
also	O
known	O
as	O
NV10	B-Device
)	O
was	O
the	O
first	O
consumer-level	O
card	O
with	O
hardware-accelerated	B-General_Concept
TL	O
;	O
professional	O
3D	O
cards	O
already	O
had	O
this	O
capability	O
.	O
</s>
<s>
Hardware	B-Algorithm
transform	I-Algorithm
and	I-Algorithm
lighting	I-Algorithm
—	O
existing	O
features	O
of	O
OpenGL	B-Application
—	O
came	O
to	O
consumer-level	O
hardware	B-Architecture
in	O
the	O
'	O
90s	O
and	O
set	O
the	O
precedent	O
for	O
later	O
pixel	B-Algorithm
shader	O
and	O
vertex	O
shader	O
units	O
which	O
were	O
far	O
more	O
flexible	O
and	O
programmable	O
.	O
</s>
<s>
Nvidia	O
was	O
first	O
to	O
produce	O
a	O
chip	O
capable	O
of	O
programmable	O
shading	O
:	O
the	O
GeForce	B-Application
3	O
(	O
code	O
named	O
NV20	O
)	O
.	O
</s>
<s>
Each	O
pixel	B-Algorithm
could	O
now	O
be	O
processed	O
by	O
a	O
short	O
program	O
that	O
could	O
include	O
additional	O
image	O
textures	O
as	O
inputs	O
,	O
and	O
each	O
geometric	O
vertex	O
could	O
likewise	O
be	O
processed	O
by	O
a	O
short	O
program	O
before	O
it	O
was	O
projected	O
onto	O
the	O
screen	O
.	O
</s>
<s>
Used	O
in	O
the	O
Xbox	B-Device
console	O
,	O
this	O
chip	O
competed	O
with	O
the	O
one	O
in	O
the	O
PlayStation	B-Device
2	I-Device
,	O
which	O
used	O
a	O
custom	O
vector	O
unit	O
for	O
hardware	B-General_Concept
accelerated	I-General_Concept
vertex	O
processing	O
(	O
commonly	O
referred	O
to	O
as	O
VU0/VU1	O
)	O
.	O
</s>
<s>
The	O
earliest	O
incarnations	O
of	O
shader	O
execution	O
engines	O
used	O
in	O
Xbox	B-Device
were	O
not	O
general	O
purpose	O
and	O
could	O
not	O
execute	O
arbitrary	O
pixel	B-Algorithm
code	O
.	O
</s>
<s>
Vertices	O
and	O
pixels	B-Algorithm
were	O
processed	O
by	O
different	O
units	O
which	O
had	O
their	O
own	O
resources	O
,	O
with	O
pixel	B-Algorithm
shaders	O
having	O
much	O
tighter	O
constraints	O
(	O
because	O
they	O
execute	O
at	O
much	O
higher	O
frequencies	O
than	O
vertices	O
)	O
.	O
</s>
<s>
Pixel	B-Algorithm
shading	O
engines	O
were	O
actually	O
more	O
akin	O
to	O
a	O
highly	O
customizable	O
function	O
block	O
and	O
did	O
not	O
really	O
"	O
run	O
"	O
a	O
program	O
.	O
</s>
<s>
Many	O
of	O
these	O
disparities	O
between	O
vertex	O
and	O
pixel	B-Algorithm
shading	O
were	O
not	O
addressed	O
until	O
the	O
Unified	O
Shader	O
Model	O
.	O
</s>
<s>
In	O
October	O
2002	O
,	O
with	O
the	O
introduction	O
of	O
the	O
ATI	B-Device
Radeon	I-Device
9700	I-Device
(	O
also	O
known	O
as	O
R300	B-Device
)	O
,	O
the	O
world	O
's	O
first	O
Direct3D	B-Application
9.0	O
accelerator	O
,	O
pixel	B-Algorithm
and	O
vertex	O
shaders	O
could	O
implement	O
looping	O
and	O
lengthy	O
floating	B-Algorithm
point	I-Algorithm
math	I-Algorithm
,	O
and	O
were	O
quickly	O
becoming	O
as	O
flexible	O
as	O
CPUs	B-General_Concept
,	O
yet	O
orders	O
of	O
magnitude	O
faster	O
for	O
image-array	O
operations	O
.	O
</s>
<s>
Pixel	B-Algorithm
shading	O
is	O
often	O
used	O
for	O
bump	O
mapping	O
,	O
which	O
adds	O
texture	O
to	O
make	O
an	O
object	O
look	O
shiny	O
,	O
dull	O
,	O
rough	O
,	O
or	O
even	O
round	O
or	O
extruded	O
.	O
</s>
<s>
With	O
the	O
introduction	O
of	O
the	O
Nvidia	B-Application
GeForce	I-Application
8	O
series	O
and	O
new	O
generic	O
stream	B-Application
processing	I-Application
units	O
,	O
GPUs	B-Architecture
became	O
more	O
generalized	O
computing	O
devices	O
.	O
</s>
<s>
parallel	B-Operating_System
GPUs	B-Architecture
are	O
making	O
computational	O
inroads	O
against	O
the	O
CPU	O
,	O
and	O
a	O
subfield	O
of	O
research	O
,	O
dubbed	O
GPU	B-Architecture
computing	I-Architecture
or	O
GPGPU	B-Architecture
for	O
general	O
purpose	O
computing	O
on	O
GPU	B-Architecture
,	O
has	O
found	O
applications	O
in	O
fields	O
as	O
diverse	O
as	O
machine	O
learning	O
,	O
oil	O
exploration	O
,	O
scientific	O
image	B-Algorithm
processing	I-Algorithm
,	O
linear	B-Language
algebra	I-Language
,	O
statistics	O
,	O
3D	B-Algorithm
reconstruction	I-Algorithm
,	O
and	O
stock	O
options	O
pricing	O
.	O
</s>
<s>
GPGPU	B-Architecture
was	O
the	O
precursor	O
to	O
what	O
is	O
now	O
called	O
a	O
compute	B-Operating_System
shader	I-Operating_System
(	O
e.g.	O
</s>
<s>
CUDA	B-Architecture
,	O
OpenCL	B-Application
,	O
DirectCompute	O
)	O
and	O
actually	O
abused	O
the	O
hardware	B-Architecture
to	O
a	O
degree	O
by	O
treating	O
the	O
data	O
passed	O
to	O
algorithms	O
as	O
texture	O
maps	O
and	O
executing	O
algorithms	O
by	O
drawing	O
a	O
triangle	O
or	O
quad	O
with	O
an	O
appropriate	O
pixel	B-Algorithm
shader	O
.	O
</s>
<s>
This	O
obviously	O
entails	O
some	O
overheads	O
since	O
units	O
like	O
the	O
scan	O
converter	O
are	O
involved	O
where	O
they	O
are	O
not	O
needed	O
(	O
nor	O
are	O
triangle	O
manipulations	O
even	O
a	O
concern	O
—	O
except	O
to	O
invoke	O
the	O
pixel	B-Algorithm
shader	O
)	O
.	O
</s>
<s>
Nvidia	O
's	O
CUDA	B-Architecture
platform	I-Architecture
,	O
first	O
introduced	O
in	O
2007	O
,	O
was	O
the	O
earliest	O
widely	O
adopted	O
programming	O
model	O
for	O
GPU	B-Architecture
computing	I-Architecture
.	O
</s>
<s>
OpenCL	B-Application
is	O
an	O
open	O
standard	O
defined	O
by	O
the	B-Library
Khronos	I-Library
Group	I-Library
that	O
allows	O
for	O
the	O
development	O
of	O
code	O
for	O
both	O
GPUs	B-Architecture
and	O
CPUs	B-General_Concept
with	O
an	O
emphasis	O
on	O
portability	O
.	O
</s>
<s>
OpenCL	B-Application
solutions	O
are	O
supported	O
by	O
Intel	O
,	O
AMD	O
,	O
Nvidia	O
,	O
and	O
ARM	O
,	O
and	O
according	O
to	O
a	O
recent	O
report	O
by	O
Evan	O
's	O
Data	O
,	O
OpenCL	B-Application
is	O
the	O
GPGPU	B-Architecture
development	O
platform	O
most	O
widely	O
used	O
by	O
developers	O
in	O
both	O
the	O
US	O
and	O
Asia	O
–	O
Pacific	O
.	O
</s>
<s>
In	O
2010	O
,	O
Nvidia	O
partnered	O
with	O
Audi	O
to	O
power	O
their	O
cars	O
 '	O
dashboards	O
,	O
using	O
the	O
Tegra	B-Operating_System
GPU	B-Architecture
to	O
provide	O
increased	O
functionality	O
to	O
cars	O
 '	O
navigation	O
and	O
entertainment	O
systems	O
.	O
</s>
<s>
Advances	O
in	O
GPU	B-Architecture
technology	O
in	O
cars	O
helped	O
advance	O
self-driving	O
technology	O
.	O
</s>
<s>
AMD	O
's	O
Radeon	B-Device
HD	I-Device
6000	O
Series	O
cards	O
were	O
released	O
in	O
2010	O
,	O
and	O
in	O
2011	O
AMD	O
released	O
its	O
6000M	O
Series	O
discrete	B-Device
GPUs	I-Device
for	O
mobile	O
devices	O
.	O
</s>
<s>
The	O
Kepler	B-General_Concept
line	O
of	O
graphics	B-Device
cards	I-Device
by	O
Nvidia	O
came	O
out	O
in	O
2012	O
and	O
were	O
used	O
in	O
the	O
Nvidia	O
's	O
600	O
and	O
700	O
series	O
cards	O
.	O
</s>
<s>
A	O
feature	O
in	O
this	O
new	O
GPU	B-Architecture
microarchitecture	O
included	O
GPU	B-Architecture
boost	O
,	O
a	O
technology	O
that	O
adjusts	O
the	O
clock-speed	O
of	O
a	O
video	B-Device
card	I-Device
to	O
increase	O
or	O
decrease	O
it	O
according	O
to	O
its	O
power	O
draw	O
.	O
</s>
<s>
The	O
Kepler	B-General_Concept
microarchitecture	I-General_Concept
was	O
manufactured	O
.	O
</s>
<s>
The	O
PS4	B-Architecture
and	O
Xbox	B-Device
One	I-Device
were	O
released	O
in	O
2013	O
;	O
they	O
both	O
use	O
GPUs	B-Architecture
based	O
on	O
AMD	O
's	O
Radeon	B-Device
HD	I-Device
7850	O
and	O
7790	O
.	O
</s>
<s>
Nvidia	O
's	O
Kepler	B-General_Concept
line	O
of	O
GPUs	B-Architecture
was	O
followed	O
by	O
the	O
Maxwell	B-General_Concept
line	O
,	O
manufactured	O
on	O
the	O
same	O
process	O
.	O
</s>
<s>
Virtual	B-Device
reality	I-Device
headsets	I-Device
have	O
high	O
system	O
requirements	O
;	O
manufacturers	O
recommended	O
the	O
GTX	O
970	O
and	O
the	O
R9	O
290X	O
or	O
better	O
at	O
the	O
time	O
of	O
their	O
release	O
.	O
</s>
<s>
Cards	O
based	O
on	O
the	O
Pascal	B-General_Concept
microarchitecture	I-General_Concept
were	O
released	O
in	O
2016	O
.	O
</s>
<s>
The	O
GeForce	B-Device
10	I-Device
series	I-Device
of	O
cards	O
are	O
under	O
this	O
generation	O
of	O
graphics	B-Device
cards	I-Device
.	O
</s>
<s>
Nvidia	O
released	O
one	O
non-consumer	O
card	O
under	O
the	O
new	O
Volta	B-General_Concept
architecture	O
,	O
the	O
Titan	O
V	O
.	O
Changes	O
from	O
the	O
Titan	B-Device
XP	I-Device
,	O
Pascal	B-General_Concept
's	O
high-end	O
card	O
,	O
include	O
an	O
increase	O
in	O
the	O
number	O
of	O
CUDA	B-Architecture
cores	O
,	O
the	O
addition	O
of	O
tensor	O
cores	O
,	O
and	O
HBM2	O
.	O
</s>
<s>
Tensor	O
cores	O
are	O
designed	O
for	O
deep	B-Algorithm
learning	I-Algorithm
,	O
while	O
high-bandwidth	O
memory	B-General_Concept
is	O
on-die	O
,	O
stacked	O
,	O
lower-clocked	O
memory	B-General_Concept
that	O
offers	O
an	O
extremely	O
wide	O
memory	B-General_Concept
bus	O
.	O
</s>
<s>
To	O
emphasize	O
that	O
the	O
Titan	O
V	O
is	O
not	O
a	O
gaming	O
card	O
,	O
Nvidia	O
removed	O
the	O
"	O
GeForce	B-Application
GTX	I-Application
"	O
suffix	O
it	O
adds	O
to	O
consumer	O
gaming	O
cards	O
.	O
</s>
<s>
In	O
2018	O
,	O
Nvidia	O
launched	O
the	O
RTX	B-Algorithm
20	O
series	O
GPUs	B-Architecture
that	O
added	O
ray-tracing	B-Algorithm
cores	O
to	O
GPUs	B-Architecture
,	O
improving	O
their	O
performance	O
on	O
lighting	O
effects	O
.	O
</s>
<s>
Polaris	B-Device
11	I-Device
and	O
Polaris	B-Device
10	I-Device
GPUs	B-Architecture
from	O
AMD	O
are	O
fabricated	O
by	O
a	O
14nm	O
process	O
.	O
</s>
<s>
Their	O
release	O
resulted	O
in	O
a	O
substantial	O
increase	O
in	O
the	O
performance	O
per	O
watt	O
of	O
AMD	O
video	B-Device
cards	I-Device
.	O
</s>
<s>
AMD	O
also	O
released	O
the	O
Vega	O
GPU	B-Architecture
series	O
for	O
the	O
high	O
end	O
market	O
as	O
a	O
competitor	O
to	O
Nvidia	O
's	O
high	O
end	O
Pascal	B-General_Concept
cards	O
,	O
also	O
featuring	O
HBM2	O
like	O
the	O
Titan	O
V	O
.	O
</s>
<s>
In	O
2019	O
,	O
AMD	O
released	O
the	O
successor	O
to	O
their	O
Graphics	B-Architecture
Core	I-Architecture
Next	I-Architecture
(	O
GCN	O
)	O
microarchitecture/instruction	O
set	O
.	O
</s>
<s>
Dubbed	O
RDNA	O
,	O
the	O
first	O
product	O
featuring	O
it	O
was	O
the	O
Radeon	B-Device
RX	I-Device
5000	I-Device
series	I-Device
of	O
video	B-Device
cards	I-Device
.	O
</s>
<s>
AMD	O
unveiled	O
the	O
Radeon	B-Device
RX	I-Device
6000	I-Device
series	I-Device
,	O
its	O
RDNA	O
2	O
graphics	B-Device
cards	I-Device
with	O
support	O
for	O
hardware-accelerated	B-General_Concept
ray	B-Algorithm
tracing	I-Algorithm
.	O
</s>
<s>
The	O
lineup	O
consisted	O
of	O
the	O
RX	B-Device
6800	I-Device
,	O
RX	B-Device
6800	I-Device
XT	I-Device
,	O
and	O
RX	B-Device
6900	I-Device
XT	I-Device
.	O
</s>
<s>
The	O
RX	B-Device
6800	I-Device
and	O
6800	O
XT	O
launched	O
on	O
November	O
18	O
,	O
2020	O
;	O
the	O
RX	B-Device
6900	I-Device
XT	I-Device
on	O
December	O
8	O
,	O
2020	O
.	O
</s>
<s>
The	O
PlayStation	B-Device
5	I-Device
and	O
Xbox	B-Device
Series	I-Device
X	I-Device
and	I-Device
Series	I-Device
S	I-Device
were	O
released	O
in	O
2020	O
;	O
they	O
both	O
use	O
GPUs	B-Architecture
based	O
on	O
the	O
RDNA	O
2	O
microarchitecture	O
with	O
proprietary	B-Application
tweaks	O
and	O
different	O
GPU	B-Architecture
configurations	O
in	O
each	O
system	O
's	O
implementation	O
.	O
</s>
<s>
Intel	O
first	O
entered	B-Device
the	I-Device
GPU	I-Device
market	I-Device
in	O
the	O
late	O
1990s	O
,	O
but	O
produced	O
lackluster	O
3D	B-Architecture
accelerators	I-Architecture
compared	O
to	O
the	O
competition	O
at	O
the	O
time	O
.	O
</s>
<s>
Rather	O
than	O
attempting	O
to	O
compete	O
with	O
the	O
high-end	O
manufacturers	O
Nvidia	O
and	O
ATI/AMD	O
,	O
they	O
began	O
integrating	O
Intel	B-Application
Graphics	I-Application
Technology	I-Application
GPUs	B-Architecture
into	O
motherboard	B-Device
chipsets	O
,	O
beginning	O
with	O
the	O
Intel	B-Device
810	I-Device
for	O
the	O
Pentium	O
III	O
,	O
and	O
later	O
into	O
their	O
CPUs	B-General_Concept
,	O
beginning	O
with	O
the	O
Intel	O
Atom	O
'	O
Pineview	O
 '	O
laptop	O
processor	O
in	O
2009	O
and	O
moving	O
into	O
desktop	O
processors	O
beginning	O
with	O
the	O
first	O
generation	O
of	O
the	O
Intel	B-Device
Core	I-Device
line	O
,	O
and	O
contemporary	O
Pentiums	O
and	O
Celerons	O
,	O
in	O
2010	O
.	O
</s>
<s>
This	O
resulted	O
in	O
a	O
large	O
nominal	O
market	O
share	O
as	O
the	O
majority	O
of	O
computers	O
with	O
an	O
Intel	O
CPU	O
also	O
featured	O
this	O
embedded	O
graphics	B-Architecture
processor	I-Architecture
.	O
</s>
<s>
Intel	O
re-entered	O
the	O
discrete	B-Device
GPU	I-Device
market	O
in	O
2022	O
with	O
its	O
Arc	B-Device
series	O
which	O
competed	O
with	O
then-current	O
GeForce	B-Application
30	O
series	O
and	O
Radeon	B-Device
6000	O
series	O
cards	O
at	O
competitive	O
prices	O
.	O
</s>
<s>
Many	O
companies	O
have	O
produced	O
GPUs	B-Architecture
under	O
a	O
number	O
of	O
brand	O
names	O
.	O
</s>
<s>
However	O
,	O
those	O
numbers	O
include	O
Intel	O
's	O
integrated	O
graphics	O
solutions	O
as	O
GPUs	B-Architecture
.	O
</s>
<s>
In	O
addition	O
,	O
Matrox	B-Application
produces	O
GPUs	B-Architecture
.	O
</s>
<s>
Modern	O
smartphones	O
use	O
mostly	O
Adreno	B-General_Concept
GPUs	B-Architecture
from	O
Qualcomm	O
,	O
PowerVR	B-Device
GPUs	B-Architecture
from	O
Imagination	O
Technologies	O
,	O
and	O
Mali	B-General_Concept
GPUs	I-General_Concept
from	O
ARM	O
.	O
</s>
<s>
Modern	O
GPUs	B-Architecture
use	O
most	O
of	O
their	O
transistors	B-Application
to	O
do	O
calculations	O
related	O
to	O
3D	O
computer	O
graphics	O
.	O
</s>
<s>
In	O
addition	O
to	O
the	O
3D	O
hardware	B-Architecture
,	O
today	O
's	O
GPUs	B-Architecture
include	O
basic	O
2D	O
acceleration	O
and	O
framebuffer	B-Algorithm
capabilities	O
(	O
usually	O
with	O
a	O
VGA	B-Protocol
compatibility	O
mode	O
)	O
.	O
</s>
<s>
Newer	O
cards	O
such	O
as	O
AMD/ATI	O
HD5000	O
–	O
HD7000	O
lack	O
dedicated	O
2D	O
acceleration	O
;	O
it	O
has	O
to	O
be	O
emulated	O
by	O
3D	O
hardware	B-Architecture
.	O
</s>
<s>
GPUs	B-Architecture
were	O
initially	O
used	O
to	O
accelerate	O
the	O
memory-intensive	O
work	O
of	O
texture	O
mapping	O
and	O
rendering	O
polygons	O
.	O
</s>
<s>
They	O
later	O
added	O
to	O
accelerate	O
geometric	O
calculations	O
such	O
as	O
the	O
rotation	O
and	O
translation	B-Algorithm
of	O
vertices	O
into	O
different	O
coordinate	O
systems	O
.	O
</s>
<s>
Recent	O
developments	O
in	O
GPUs	B-Architecture
include	O
support	O
for	O
programmable	O
shaders	O
which	O
can	O
manipulate	O
vertices	O
and	O
textures	O
with	O
many	O
of	O
the	O
same	O
operations	O
that	O
are	O
supported	O
by	O
CPUs	B-General_Concept
,	O
oversampling	O
and	O
interpolation	B-Algorithm
techniques	O
to	O
reduce	O
aliasing	B-Error_Name
,	O
and	O
very	O
high-precision	O
color	B-Algorithm
spaces	I-Algorithm
.	O
</s>
<s>
Several	O
factors	O
of	O
GPU	B-Architecture
construction	O
affect	O
the	O
performance	O
of	O
the	O
card	O
for	O
real-time	B-General_Concept
rendering	I-General_Concept
,	O
such	O
as	O
the	O
size	O
of	O
the	O
connector	O
pathways	O
in	O
the	O
semiconductor	B-Architecture
device	I-Architecture
fabrication	I-Architecture
,	O
the	O
clock	O
signal	O
frequency	O
,	O
and	O
the	O
number	O
and	O
size	O
of	O
various	O
on-chip	O
memory	B-General_Concept
caches	B-General_Concept
.	O
</s>
<s>
Additionally	O
,	O
the	O
number	O
of	O
Streaming	O
Multiprocessors	O
(	O
SM	O
)	O
for	O
NVidia	B-Architecture
GPUs	I-Architecture
,	O
or	O
Compute	O
Units	O
(	O
CU	O
)	O
for	O
AMD	B-Architecture
GPUs	I-Architecture
,	O
which	O
describe	O
the	O
number	O
of	O
core	O
on-silicon	O
processor	O
units	O
within	O
the	O
GPU	B-Architecture
chip	O
that	O
perform	O
the	O
core	O
calculations	O
,	O
typically	O
working	O
in	O
parallel	B-Operating_System
with	O
other	O
SM/CUs	O
on	O
the	O
GPU	B-Architecture
.	O
</s>
<s>
Performance	O
of	O
GPUs	B-Architecture
is	O
typically	O
measured	O
in	O
floating	B-Algorithm
point	I-Algorithm
operations	O
per	O
second	O
or	O
FLOPS	O
;	O
GPUs	B-Architecture
in	O
the	O
2010s	O
and	O
2020s	O
typically	O
deliver	O
performance	O
measured	O
in	O
teraflops	O
(	O
TFLOPS	O
)	O
.	O
</s>
<s>
Most	O
GPUs	B-Architecture
made	O
since	O
1995	O
support	O
the	O
YUV	O
color	B-Algorithm
space	I-Algorithm
and	O
hardware	B-Architecture
overlays	I-Architecture
,	O
important	O
for	O
digital	O
video	O
playback	O
,	O
and	O
many	O
GPUs	B-Architecture
made	O
since	O
2000	O
also	O
support	O
MPEG	O
primitives	O
such	O
as	O
motion	B-Algorithm
compensation	I-Algorithm
and	O
iDCT	B-General_Concept
.	O
</s>
<s>
This	O
hardware-accelerated	B-General_Concept
video	B-Application
decoding	I-Application
,	O
in	O
which	O
portions	O
of	O
the	O
video	B-Application
decoding	I-Application
process	O
and	O
video	O
post-processing	O
are	O
offloaded	O
to	O
the	O
GPU	B-Architecture
hardware	B-Architecture
,	O
is	O
commonly	O
referred	O
to	O
as	O
"	O
GPU	B-Architecture
accelerated	O
video	B-Application
decoding	I-Application
"	O
,	O
"	O
GPU	B-Architecture
assisted	O
video	B-Application
decoding	I-Application
"	O
,	O
"	O
GPU	B-Architecture
hardware	B-General_Concept
accelerated	I-General_Concept
video	B-Application
decoding	I-Application
"	O
,	O
or	O
"	O
GPU	B-Architecture
hardware	B-Architecture
assisted	O
video	B-Application
decoding	I-Application
"	O
.	O
</s>
<s>
Recent	O
graphics	B-Device
cards	I-Device
decode	O
high-definition	O
video	O
on	O
the	O
card	O
,	O
offloading	O
the	O
central	B-General_Concept
processing	I-General_Concept
unit	I-General_Concept
.	O
</s>
<s>
The	O
most	O
common	O
APIs	B-General_Concept
for	O
GPU	B-Architecture
accelerated	O
video	B-Application
decoding	I-Application
are	O
DxVA	B-General_Concept
for	O
Microsoft	B-Application
Windows	I-Application
operating	I-Application
system	I-Application
and	O
VDPAU	B-Application
,	O
VAAPI	B-Application
,	O
XvMC	B-Protocol
,	O
and	O
XvBA	B-Application
for	O
Linux-based	O
and	O
UNIX-like	O
operating	O
systems	O
.	O
</s>
<s>
All	O
except	O
XvMC	B-Protocol
are	O
capable	O
of	O
decoding	O
videos	O
encoded	O
with	O
MPEG-1	B-Algorithm
,	O
MPEG-2	B-Algorithm
,	O
MPEG-4	B-Application
ASP	I-Application
(	O
MPEG-4	B-Application
Part	I-Application
2	I-Application
)	O
,	O
MPEG-4	B-Application
AVC	I-Application
(	O
H.264	B-Application
/	O
DivX	B-Application
6	O
)	O
,	O
VC-1	B-Operating_System
,	O
WMV3/WMV9	O
,	O
Xvid	B-Algorithm
/	O
OpenDivX	B-Algorithm
(	O
DivX	B-Application
4	O
)	O
,	O
and	O
DivX	B-Application
5	O
codecs	B-General_Concept
,	O
while	O
XvMC	B-Protocol
is	O
only	O
capable	O
of	O
decoding	O
MPEG-1	B-Algorithm
and	O
MPEG-2	B-Algorithm
.	O
</s>
<s>
There	O
are	O
several	O
dedicated	O
hardware	B-Architecture
video	B-Application
decoding	I-Application
and	O
encoding	O
solutions	O
.	O
</s>
<s>
Video	B-Application
decoding	I-Application
processes	O
that	O
can	O
be	O
accelerated	O
by	O
modern	O
GPU	B-Architecture
hardware	B-Architecture
are	O
:	O
</s>
<s>
Given	O
that	O
most	O
of	O
these	O
computations	O
involve	O
matrix	B-Architecture
and	O
vector	O
operations	O
,	O
engineers	O
and	O
scientists	O
have	O
increasingly	O
studied	O
the	O
use	O
of	O
GPUs	B-Architecture
for	O
non-graphical	O
calculations	O
.	O
</s>
<s>
They	O
are	O
especially	O
suited	O
to	O
other	O
embarrassingly	B-Operating_System
parallel	I-Operating_System
problems	I-Operating_System
.	O
</s>
<s>
One	O
notable	O
use	O
of	O
GPUs	B-Architecture
is	O
in	O
neural	O
networks	O
 '	O
training	O
.	O
</s>
<s>
In	O
research	O
done	O
by	O
Indigo	O
,	O
it	O
was	O
found	O
that	O
while	O
training	O
deep	B-Algorithm
learning	I-Algorithm
neural	O
networks	O
,	O
GPUs	B-Architecture
can	O
be	O
250	O
times	O
faster	O
than	O
CPUs	B-General_Concept
.	O
</s>
<s>
In	O
the	O
1970s	O
,	O
the	O
term	O
"	O
GPU	B-Architecture
"	O
originally	O
stood	O
for	O
graphics	B-Architecture
processor	I-Architecture
unit	O
and	O
described	O
a	O
programmable	O
processing	O
unit	O
working	O
independently	O
from	O
the	O
CPU	O
that	O
was	O
responsible	O
for	O
graphics	O
manipulation	O
and	O
output	O
.	O
</s>
<s>
In	O
1994	O
,	O
Sony	O
used	O
the	O
term	O
(	O
now	O
standing	O
for	O
graphics	B-Architecture
processing	I-Architecture
unit	I-Architecture
)	O
in	O
reference	O
to	O
the	O
PlayStation	B-Device
console	I-Device
's	O
Toshiba-designed	O
Sony	B-Device
GPU	I-Device
.	O
</s>
<s>
The	O
term	O
was	O
popularized	O
by	O
Nvidia	O
in	O
1999	O
,	O
who	O
marketed	O
the	O
GeForce	B-Device
256	I-Device
as	O
"	O
the	O
world	O
's	O
first	O
GPU	B-Architecture
"	O
.	O
</s>
<s>
It	O
was	O
presented	O
as	O
a	O
"	O
single-chip	O
processor	O
with	O
integrated	O
transform	B-Algorithm
,	I-Algorithm
lighting	I-Algorithm
,	I-Algorithm
triangle	I-Algorithm
setup/clipping	I-Algorithm
,	O
and	O
rendering	O
engines	O
"	O
.	O
</s>
<s>
Rival	O
ATI	O
Technologies	O
coined	O
the	O
term	O
"	O
visual	B-Architecture
processing	I-Architecture
unit	I-Architecture
"	O
or	O
VPU	O
with	O
the	O
release	O
of	O
the	O
Radeon	B-Device
9700	I-Device
in	O
2002	O
.	O
</s>
<s>
In	O
personal	B-Device
computers	I-Device
,	O
there	O
are	O
two	O
main	O
forms	O
of	O
GPUs	B-Architecture
.	O
</s>
<s>
Integrated	O
graphics	O
-	O
also	O
called	O
shared	O
graphics	O
solutions	O
,	O
integrated	O
graphics	B-Architecture
processors	I-Architecture
(	O
IGP	O
)	O
,	O
or	O
unified	O
memory	B-General_Concept
architecture	O
(	O
UMA	O
)	O
.	O
</s>
<s>
Most	O
GPUs	B-Architecture
are	O
designed	O
for	O
a	O
specific	O
use	O
,	O
real-time	B-General_Concept
3D	O
graphics	O
,	O
or	O
other	O
mass	O
calculations	O
:	O
</s>
<s>
Dedicated	O
graphics	B-Architecture
processing	I-Architecture
units	I-Architecture
are	O
not	O
necessarily	O
removable	O
,	O
nor	O
does	O
it	O
necessarily	O
interface	O
with	O
the	O
motherboard	B-Device
in	O
a	O
standard	O
fashion	O
.	O
</s>
<s>
The	O
term	O
"	O
dedicated	O
"	O
refers	O
to	O
the	O
fact	O
that	O
graphics	B-Device
cards	I-Device
have	O
RAM	B-Architecture
that	O
is	O
dedicated	O
to	O
the	O
card	O
's	O
use	O
,	O
not	O
to	O
the	O
fact	O
that	O
most	O
dedicated	O
GPUs	B-Architecture
are	O
removable	O
.	O
</s>
<s>
This	O
RAM	B-Architecture
is	O
usually	O
specially	O
selected	O
for	O
the	O
expected	O
serial	O
workload	O
of	O
the	O
graphics	B-Device
card	I-Device
(	O
see	O
GDDR	B-Device
)	O
.	O
</s>
<s>
Sometimes	O
,	O
systems	O
with	O
dedicated	O
,	O
discrete	B-Device
GPUs	I-Device
were	O
called	O
"	O
DIS	O
"	O
systems	O
,	O
as	O
opposed	O
to	O
"	O
UMA	O
"	O
systems	O
(	O
see	O
next	O
section	O
)	O
.	O
</s>
<s>
Dedicated	O
GPUs	B-Architecture
for	O
portable	O
computers	O
are	O
most	O
commonly	O
interfaced	O
through	O
a	O
non-standard	O
and	O
often	O
proprietary	B-Application
slot	O
due	O
to	O
size	O
and	O
weight	O
constraints	O
.	O
</s>
<s>
Graphics	B-Device
cards	I-Device
with	O
dedicated	O
GPUs	B-Architecture
typically	O
interface	O
with	O
the	O
motherboard	B-Device
by	O
means	O
of	O
an	O
expansion	B-Device
slot	I-Device
such	O
as	O
PCI	B-Protocol
Express	O
(	O
PCIe	O
)	O
or	O
Accelerated	B-Architecture
Graphics	I-Architecture
Port	I-Architecture
(	O
AGP	O
)	O
.	O
</s>
<s>
They	O
can	O
usually	O
be	O
replaced	O
or	O
upgraded	O
with	O
relative	O
ease	O
,	O
assuming	O
the	O
motherboard	B-Device
is	O
capable	O
of	O
supporting	O
the	O
upgrade	O
.	O
</s>
<s>
A	O
few	O
graphics	B-Device
cards	I-Device
still	O
use	O
Peripheral	B-Protocol
Component	I-Protocol
Interconnect	I-Protocol
(	O
PCI	B-Protocol
)	O
slots	O
,	O
but	O
their	O
bandwidth	O
is	O
so	O
limited	O
that	O
they	O
are	O
generally	O
used	O
only	O
when	O
a	O
PCIe	O
or	O
AGP	B-Architecture
slot	I-Architecture
is	O
not	O
available	O
.	O
</s>
<s>
Technologies	O
such	O
as	O
SLI	B-Device
and	O
NVLink	O
by	O
Nvidia	O
and	O
CrossFire	O
by	O
AMD	O
allow	O
multiple	O
GPUs	B-Architecture
to	O
draw	O
images	B-Algorithm
simultaneously	O
for	O
a	O
single	O
screen	O
,	O
increasing	O
the	O
processing	O
power	O
available	O
for	O
graphics	O
.	O
</s>
<s>
These	O
technologies	O
,	O
however	O
,	O
are	O
increasingly	O
uncommon	O
;	O
most	O
games	O
do	O
not	O
fully	O
utilize	O
multiple	O
GPUs	B-Architecture
,	O
as	O
most	O
users	O
cannot	O
afford	O
them	O
.	O
</s>
<s>
Multiple	O
GPUs	B-Architecture
are	O
still	O
used	O
on	O
supercomputers	O
(	O
like	O
in	O
Summit	B-Device
)	O
,	O
on	O
workstations	B-Device
to	O
accelerate	O
video	O
(	O
processing	O
multiple	O
videos	O
at	O
once	O
)	O
and	O
3D	O
rendering	O
,	O
for	O
VFX	O
and	O
for	O
simulations	O
,	O
and	O
in	O
AI	O
to	O
expedite	O
training	O
,	O
as	O
is	O
the	O
case	O
with	O
Nvidia	O
's	O
lineup	O
of	O
DGX	O
workstations	B-Device
and	O
servers	O
,	O
Tesla	O
GPUs	B-Architecture
,	O
and	O
Intel	O
's	O
Ponte	O
Vecchio	O
GPUs	B-Architecture
.	O
</s>
<s>
Integrated	O
graphics	B-Architecture
processing	I-Architecture
unit	I-Architecture
(	O
IGPU	O
)	O
,	O
Integrated	O
graphics	O
,	O
shared	O
graphics	O
solutions	O
,	O
integrated	O
graphics	B-Architecture
processors	I-Architecture
(	O
IGP	O
)	O
,	O
or	O
unified	O
memory	B-General_Concept
architecture	O
(	O
UMA	O
)	O
utilize	O
a	O
portion	O
of	O
a	O
computer	O
's	O
system	O
RAM	B-Architecture
rather	O
than	O
dedicated	O
graphics	O
memory	B-General_Concept
.	O
</s>
<s>
IGPs	O
can	O
be	O
integrated	O
onto	O
the	O
motherboard	B-Device
as	O
part	O
of	O
the	O
(	O
northbridge	B-Device
)	O
chipset	O
,	O
or	O
on	O
the	O
same	O
die	O
(	O
integrated	O
circuit	O
)	O
with	O
the	O
CPU	O
(	O
like	O
AMD	B-Architecture
APU	I-Architecture
or	O
Intel	B-Application
HD	I-Application
Graphics	I-Application
)	O
.	O
</s>
<s>
On	O
certain	O
motherboards	B-Device
,	O
AMD	O
's	O
IGPs	O
can	O
use	O
dedicated	O
sideport	O
memory	B-General_Concept
:	O
a	O
separate	O
fixed	O
block	O
of	O
high	O
performance	O
memory	B-General_Concept
that	O
is	O
dedicated	O
for	O
use	O
by	O
the	O
GPU	B-Architecture
.	O
</s>
<s>
However	O
,	O
modern	O
integrated	O
graphics	B-Architecture
processors	I-Architecture
such	O
as	O
AMD	B-Architecture
Accelerated	I-Architecture
Processing	I-Architecture
Unit	I-Architecture
and	O
Intel	B-Application
Graphics	I-Application
Technology	I-Application
(	O
HD	O
,	O
UHD	O
,	O
Iris	O
,	O
Iris	O
Pro	O
,	O
Iris	O
Plus	O
,	O
and	O
Xe-LP	O
)	O
can	O
handle	O
2D	O
graphics	O
or	O
low	O
stress	O
3D	O
graphics	O
.	O
</s>
<s>
Since	O
GPU	B-Architecture
computations	O
are	O
memory-intensive	O
,	O
integrated	O
processing	O
may	O
compete	O
with	O
the	O
CPU	O
for	O
relatively	O
slow	O
system	O
RAM	B-Architecture
,	O
as	O
it	O
has	O
minimal	O
or	O
no	O
dedicated	O
video	O
memory	B-General_Concept
.	O
</s>
<s>
IGPs	O
use	O
system	O
memory	B-General_Concept
with	O
bandwidth	O
up	O
to	O
a	O
current	O
maximum	O
of	O
128GB/s	O
,	O
whereas	O
a	O
discrete	B-Device
graphics	I-Device
card	O
may	O
have	O
a	O
bandwidth	O
of	O
more	O
than	O
1000GB/s	O
between	O
its	O
VRAM	O
and	O
GPU	B-Architecture
core	O
.	O
</s>
<s>
This	O
memory	B-General_Concept
bus	O
bandwidth	O
can	O
limit	O
the	O
performance	O
of	O
the	O
GPU	B-Architecture
,	O
though	O
multi-channel	B-Architecture
memory	I-Architecture
can	O
mitigate	O
this	O
deficiency	O
.	O
</s>
<s>
Older	O
integrated	O
graphics	O
chipsets	O
lacked	O
hardware	B-Algorithm
transform	I-Algorithm
and	I-Algorithm
lighting	I-Algorithm
,	O
but	O
newer	O
ones	O
include	O
it	O
.	O
</s>
<s>
Hybrid	O
GPUs	B-Architecture
compete	O
with	O
integrated	O
graphics	O
in	O
the	O
low-end	O
desktop	O
and	O
notebook	O
markets	O
.	O
</s>
<s>
Hybrid	O
graphics	B-Device
cards	I-Device
are	O
somewhat	O
more	O
expensive	O
than	O
integrated	O
graphics	O
,	O
but	O
much	O
less	O
expensive	O
than	O
dedicated	O
graphics	B-Device
cards	I-Device
.	O
</s>
<s>
They	O
share	O
memory	B-General_Concept
with	O
the	O
system	O
and	O
have	O
a	O
small	O
dedicated	O
memory	B-General_Concept
cache	O
,	O
to	O
make	O
up	O
for	O
the	O
high	O
latency	B-General_Concept
of	O
the	O
system	O
RAM	B-Architecture
.	O
</s>
<s>
Technologies	O
within	O
PCI	B-Protocol
Express	O
make	O
this	O
possible	O
.	O
</s>
<s>
While	O
these	O
solutions	O
are	O
sometimes	O
advertised	O
as	O
having	O
as	O
much	O
as	O
768MB	O
of	O
RAM	B-Architecture
,	O
this	O
refers	O
to	O
how	O
much	O
can	O
be	O
shared	O
with	O
the	O
system	O
memory	B-General_Concept
.	O
</s>
<s>
It	O
is	O
common	O
to	O
use	O
a	O
general	O
purpose	O
graphics	B-Architecture
processing	I-Architecture
unit	I-Architecture
(	O
GPGPU	B-Architecture
)	O
as	O
a	O
modified	O
form	O
of	O
stream	B-Application
processor	I-Application
(	O
or	O
a	O
vector	B-Operating_System
processor	I-Operating_System
)	O
,	O
running	O
compute	B-Operating_System
kernels	I-Operating_System
.	O
</s>
<s>
This	O
turns	O
the	O
massive	O
computational	O
power	O
of	O
a	O
modern	O
graphics	B-Architecture
accelerator	I-Architecture
's	O
shader	O
pipeline	O
into	O
general-purpose	O
computing	O
power	O
.	O
</s>
<s>
The	O
two	O
largest	O
discrete	O
(	O
see	O
"	O
Dedicated	O
graphics	B-Device
cards	I-Device
"	O
above	O
)	O
GPU	B-Architecture
designers	O
,	O
AMD	O
and	O
Nvidia	O
,	O
are	O
pursuing	O
this	O
approach	O
with	O
an	O
array	O
of	O
applications	O
.	O
</s>
<s>
Both	O
Nvidia	O
and	O
AMD	O
teamed	O
with	O
Stanford	O
University	O
to	O
create	O
a	O
GPU-based	O
client	O
for	O
the	O
Folding	B-Application
@home	I-Application
distributed	O
computing	O
project	O
for	O
protein	O
folding	O
calculations	O
.	O
</s>
<s>
In	O
certain	O
circumstances	O
,	O
the	O
GPU	B-Architecture
calculates	O
forty	O
times	O
faster	O
than	O
the	O
CPUs	B-General_Concept
traditionally	O
used	O
by	O
such	O
applications	O
.	O
</s>
<s>
GPGPU	B-Architecture
can	O
be	O
used	O
for	O
many	O
types	O
of	O
embarrassingly	B-Operating_System
parallel	I-Operating_System
tasks	O
including	O
ray	B-Algorithm
tracing	I-Algorithm
.	O
</s>
<s>
They	O
are	O
generally	O
suited	O
to	O
high-throughput	O
computations	O
that	O
exhibit	O
data-parallelism	B-Operating_System
to	O
exploit	O
the	O
wide	O
vector	O
width	O
SIMD	B-Device
architecture	O
of	O
the	O
GPU	B-Architecture
.	O
</s>
<s>
GPU-based	O
high	O
performance	O
computers	O
play	O
a	O
significant	O
role	O
in	O
large-scale	O
modelling	O
.	O
</s>
<s>
Three	O
of	O
the	O
ten	O
most	O
powerful	O
supercomputers	O
in	O
the	O
world	O
take	O
advantage	O
of	O
GPU	B-Architecture
acceleration	O
.	O
</s>
<s>
GPUs	B-Architecture
support	O
API	B-General_Concept
extensions	O
to	O
the	O
C	B-Language
programming	I-Language
language	I-Language
such	O
as	O
OpenCL	B-Application
and	O
OpenMP	B-Application
.	O
</s>
<s>
Furthermore	O
,	O
each	O
GPU	B-Architecture
vendor	O
introduced	O
its	O
own	O
API	B-General_Concept
which	O
only	O
works	O
with	O
their	O
cards	O
:	O
AMD	B-Application
APP	I-Application
SDK	I-Application
from	O
AMD	O
,	O
and	O
CUDA	B-Architecture
from	O
Nvidia	O
.	O
</s>
<s>
These	O
allow	O
functions	O
called	O
compute	B-Operating_System
kernels	I-Operating_System
to	O
run	O
on	O
the	O
GPU	B-Architecture
's	O
stream	O
processors	O
.	O
</s>
<s>
This	O
makes	O
it	O
possible	O
for	O
C	B-Language
programs	I-Language
to	O
take	O
advantage	O
of	O
a	O
GPU	B-Architecture
's	O
ability	O
to	O
operate	O
on	O
large	O
buffers	O
in	O
parallel	B-Operating_System
,	O
while	O
still	O
using	O
the	O
CPU	O
when	O
appropriate	O
.	O
</s>
<s>
CUDA	B-Architecture
was	O
the	O
first	O
API	B-General_Concept
to	O
allow	O
CPU-based	O
applications	O
to	O
directly	O
access	O
the	O
resources	O
of	O
a	O
GPU	B-Architecture
for	O
more	O
general	O
purpose	O
computing	O
without	O
the	O
limitations	O
of	O
using	O
a	O
graphics	B-Library
API	I-Library
.	O
</s>
<s>
Since	O
2005	O
there	O
has	O
been	O
interest	O
in	O
using	O
the	O
performance	O
offered	O
by	O
GPUs	B-Architecture
for	O
evolutionary	O
computation	O
in	O
general	O
,	O
and	O
for	O
accelerating	O
the	O
fitness	O
evaluation	O
in	O
genetic	B-Algorithm
programming	I-Algorithm
in	O
particular	O
.	O
</s>
<s>
Most	O
approaches	O
compile	O
linear	B-Algorithm
or	O
tree	B-Algorithm
programs	I-Algorithm
on	O
the	O
host	O
PC	O
and	O
transfer	O
the	O
executable	O
to	O
the	O
GPU	B-Architecture
to	O
be	O
run	O
.	O
</s>
<s>
Typically	O
the	O
performance	O
advantage	O
is	O
only	O
obtained	O
by	O
running	O
the	O
single	O
active	O
program	O
simultaneously	O
on	O
many	O
example	O
problems	O
in	O
parallel	B-Operating_System
,	O
using	O
the	O
GPU	B-Architecture
's	O
SIMD	B-Device
architecture	O
.	O
</s>
<s>
However	O
,	O
substantial	O
acceleration	O
can	O
also	O
be	O
obtained	O
by	O
not	O
compiling	O
the	O
programs	O
,	O
and	O
instead	O
transferring	O
them	O
to	O
the	O
GPU	B-Architecture
,	O
to	O
be	O
interpreted	O
there	O
.	O
</s>
<s>
A	O
modern	O
GPU	B-Architecture
can	O
simultaneously	O
interpret	O
hundreds	O
of	O
thousands	O
of	O
very	O
small	O
programs	O
.	O
</s>
<s>
Some	O
modern	O
workstation	B-Device
GPUs	B-Architecture
,	O
such	O
as	O
the	O
Nvidia	B-Application
Quadro	I-Application
workstation	B-Device
cards	O
using	O
the	O
Volta	B-General_Concept
and	O
Turing	O
architectures	O
,	O
feature	O
dedicating	O
processing	O
cores	O
for	O
tensor-based	O
deep	B-Algorithm
learning	I-Algorithm
applications	O
.	O
</s>
<s>
In	O
Nvidia	O
's	O
current	O
series	O
of	O
GPUs	B-Architecture
these	O
are	O
called	O
Tensor	O
Cores	O
.	O
</s>
<s>
These	O
GPUs	B-Architecture
usually	O
have	O
significant	O
FLOPS	O
performance	O
increases	O
,	O
utilizing	O
4×4	O
matrix	B-Architecture
multiplication	O
and	O
division	O
,	O
resulting	O
in	O
hardware	B-Architecture
performance	O
up	O
to	O
128	O
TFLOPS	O
in	O
some	O
applications	O
.	O
</s>
<s>
An	O
external	O
GPU	B-Architecture
is	O
a	O
graphics	B-Architecture
processor	I-Architecture
located	O
outside	O
of	O
the	O
housing	O
of	O
the	O
computer	O
,	O
similar	O
to	O
a	O
large	O
external	O
hard	O
drive	O
.	O
</s>
<s>
External	O
graphics	B-Architecture
processors	I-Architecture
are	O
sometimes	O
used	O
with	O
laptop	O
computers	O
.	O
</s>
<s>
Laptops	O
might	O
have	O
a	O
substantial	O
amount	O
of	O
RAM	B-Architecture
and	O
a	O
sufficiently	O
powerful	O
central	B-General_Concept
processing	I-General_Concept
unit	I-General_Concept
(	O
CPU	O
)	O
,	O
but	O
often	O
lack	O
a	O
powerful	O
graphics	B-Architecture
processor	I-Architecture
,	O
and	O
instead	O
have	O
a	O
less	O
powerful	O
but	O
more	O
energy-efficient	O
on-board	O
graphics	B-General_Concept
chip	I-General_Concept
.	O
</s>
<s>
On-board	O
graphics	B-General_Concept
chips	I-General_Concept
are	O
often	O
not	O
powerful	O
enough	O
for	O
playing	O
video	O
games	O
,	O
or	O
for	O
other	O
graphically	O
intensive	O
tasks	O
,	O
such	O
as	O
editing	O
video	O
or	O
3D	O
animation/rendering	O
.	O
</s>
<s>
Therefore	O
,	O
it	O
is	O
desirable	O
to	O
attach	O
a	O
GPU	B-Architecture
to	O
some	O
external	O
bus	O
of	O
a	O
notebook	O
.	O
</s>
<s>
PCI	B-Protocol
Express	O
is	O
the	O
only	O
bus	O
used	O
for	O
this	O
purpose	O
.	O
</s>
<s>
The	O
port	O
may	O
be	O
,	O
for	O
example	O
,	O
an	O
ExpressCard	B-Device
or	O
mPCIe	O
port	O
(	O
PCIe	O
×1	O
,	O
up	O
to	O
5	O
or	O
2.5	O
Gbit/s	O
respectively	O
)	O
or	O
a	O
Thunderbolt	B-Protocol
1	O
,	O
2	O
,	O
or	O
3	O
port	O
(	O
PCIe	O
×4	O
,	O
up	O
to	O
10	O
,	O
20	O
,	O
or	O
40	O
Gbit/s	O
respectively	O
)	O
.	O
</s>
<s>
eGPU	O
enclosures	O
include	O
their	O
own	O
power	O
supply	O
(	O
PSU	O
)	O
,	O
because	O
powerful	O
GPUs	B-Architecture
can	O
consume	O
hundreds	O
of	O
watts	O
.	O
</s>
<s>
Official	O
vendor	O
support	O
for	O
external	O
GPUs	B-Architecture
has	O
gained	O
traction	O
.	O
</s>
<s>
A	O
milestone	O
was	O
Apple	O
's	O
decision	O
to	O
support	O
external	O
GPUs	B-Architecture
with	O
MacOS	O
High	O
Sierra	O
10.13.4	O
.	O
</s>
<s>
Several	O
major	O
hardware	B-Architecture
vendors	O
(	O
HP	O
,	O
Alienware	O
,	O
Razer	O
)	O
released	O
Thunderbolt	B-Protocol
3	O
eGPU	O
enclosures	O
.	O
</s>
<s>
In	O
2013	O
,	O
438.3	O
million	O
GPUs	B-Architecture
were	O
shipped	O
globally	O
and	O
the	O
forecast	O
for	O
2014	O
was	O
414.2	O
million	O
.	O
</s>
