<s>
The	O
RSX	B-Device
'	I-Device
Reality	I-Device
Synthesizer	I-Device
 '	I-Device
is	O
a	O
proprietary	O
graphics	B-Architecture
processing	I-Architecture
unit	I-Architecture
(	O
GPU	B-Architecture
)	O
codeveloped	O
by	O
Nvidia	O
and	O
Sony	O
for	O
the	O
PlayStation	B-Operating_System
3	I-Operating_System
game	O
console	O
.	O
</s>
<s>
It	O
is	O
a	O
GPU	B-Architecture
based	O
on	O
the	O
Nvidia	O
7800GTX	O
graphics	B-Architecture
processor	I-Architecture
and	O
,	O
according	O
to	O
Nvidia	O
,	O
is	O
a	O
G70/G71	O
(	O
previously	O
known	O
as	O
NV47	O
)	O
hybrid	O
architecture	O
with	O
some	O
modifications	O
.	O
</s>
<s>
The	O
RSX	O
has	O
separate	O
vertex	O
and	O
pixel	O
shader	O
pipelines	B-General_Concept
.	O
</s>
<s>
The	O
GPU	B-Architecture
makes	O
use	O
of	O
256	O
MB	O
GDDR3	O
RAM	O
clocked	O
at	O
650	O
MHz	O
with	O
an	O
effective	O
transmission	O
rate	O
of	O
1.3GHz	O
and	O
up	O
to	O
224	O
MB	O
of	O
the	O
3.2GHz	O
XDR	O
main	O
memory	O
via	O
the	O
CPU	B-General_Concept
(	O
480	O
MB	O
max	O
)	O
.	O
</s>
<s>
Although	O
it	O
carries	O
the	O
majority	O
of	O
the	O
graphics	O
processing	O
,	O
the	O
Cell	B-General_Concept
Broadband	I-General_Concept
Engine	I-General_Concept
,	O
the	O
console	O
's	O
CPU	B-General_Concept
,	O
is	O
also	O
used	O
complementarily	O
for	O
some	O
graphics-related	O
computational	O
loads	O
of	O
the	O
console	O
.	O
</s>
<s>
Other	O
features	O
:	O
Support	O
for	O
Bilinear	O
,	O
trilinear	O
,	O
anisotropic	O
,	O
quincunx	O
texture	O
filtering	O
,	O
quincunx	O
antialiasing	O
,	O
up	O
to	O
4xMSAA	O
,	O
SSAA	B-Algorithm
,	O
Alpha	O
to	O
Coverage	O
and	O
Alphakill	O
.	O
</s>
<s>
65nm	B-Algorithm
:	O
</s>
<s>
The	O
4MB	O
of	O
GPU	B-Architecture
Data	O
contains	O
RAMIN	O
,	O
RAMHT	O
,	O
RAMFC	O
,	O
DMA	O
Objects	O
,	O
Graphic	O
Objects	O
,	O
and	O
the	O
Graphic	O
Context	O
.	O
</s>
<s>
The	O
shipped	O
PS3	B-Operating_System
has	O
the	O
RSX	O
clocked	O
in	O
at	O
500Mhz	O
(	O
front	O
and	O
back	O
end	O
,	O
although	O
the	O
pixel	O
shaders	O
run	O
separately	O
inside	O
at	O
550Mhz	O
)	O
.	O
</s>
<s>
Because	O
of	O
the	O
very	O
slow	O
Cell	O
Read	O
speed	O
from	O
the	O
256MB	O
GDDR3	O
memory	O
,	O
it	O
is	O
more	O
efficient	O
for	O
the	O
Cell	O
to	O
work	O
in	O
XDR	O
and	O
then	O
have	O
the	O
RSX	O
pull	O
data	O
from	O
XDR	O
and	O
write	O
to	O
GDDR3	O
for	O
output	O
to	O
the	O
HDMI	B-Protocol
display	O
.	O
</s>
<s>
The	O
easiest	O
way	O
is	O
to	O
use	O
high	O
level	O
PSGL	O
,	O
which	O
is	O
basically	O
OpenGL|ES	O
with	O
programmable	O
pipeline	O
added	O
in	O
,	O
however	O
this	O
is	O
unpopular	O
due	O
to	O
the	O
performance	O
overhead	O
on	O
a	O
relatively	O
weak	O
console	O
CPU	B-General_Concept
.	O
</s>
<s>
At	O
a	O
lower	O
level	O
developers	O
can	O
use	O
LibGCM	O
,	O
which	O
is	O
an	O
API	O
that	O
builds	O
RSX	O
command	O
buffers	B-General_Concept
at	O
a	O
lower	O
level	O
.	O
</s>
<s>
The	O
RSX	B-Device
'	I-Device
Reality	I-Device
Synthesizer	I-Device
 '	I-Device
is	O
based	O
on	O
the	O
G70	O
architecture	O
,	O
but	O
features	O
a	O
few	O
changes	O
to	O
the	O
core	O
.	O
</s>
<s>
This	O
was	O
achieved	O
by	O
enlarging	O
the	O
chip	O
size	O
to	O
accommodate	O
larger	O
buffers	B-General_Concept
and	O
caches	B-General_Concept
in	O
order	O
to	O
keep	O
the	O
graphics	B-General_Concept
pipeline	I-General_Concept
full	O
.	O
</s>
<s>
Since	O
the	O
G70	O
is	O
capable	O
of	O
carrying	O
out	O
136	O
shader	O
operations	O
per	O
clock	O
cycle	O
,	O
the	O
RSX	O
was	O
expected	O
to	O
feature	O
the	O
same	O
number	O
of	O
parallel	O
pixel	O
and	O
vertex	O
shader	O
pipelines	B-General_Concept
as	O
the	O
G70	O
,	O
which	O
contains	O
24	O
pixel	O
and	O
8	O
vertex	O
pipelines	B-General_Concept
.	O
</s>
