<s>
The	O
R420	B-Device
GPU	B-Architecture
,	O
developed	O
by	O
ATI	O
Technologies	O
,	O
was	O
the	O
company	O
's	O
basis	O
for	O
its	O
3rd-generation	O
DirectX	O
9.0/OpenGL	O
2.0-capable	O
graphics	B-Device
cards	I-Device
.	O
</s>
<s>
Used	O
first	O
on	O
the	O
Radeon	O
X800	O
,	O
the	O
R420	B-Device
was	O
produced	O
on	O
a	O
0.13	O
micrometer	O
(	O
130	O
nm	O
)	O
low-K	B-Algorithm
photolithography	B-Algorithm
process	O
and	O
used	O
GDDR-3	O
memory	O
.	O
</s>
<s>
The	O
chip	O
was	O
designed	O
for	O
AGP	B-Architecture
graphics	B-Device
cards	I-Device
.	O
</s>
<s>
Driver	O
support	O
of	O
this	O
core	O
was	O
discontinued	O
as	O
of	O
Catalyst	O
9.4	O
,	O
and	O
as	O
a	O
result	O
there	O
is	O
no	O
official	O
Windows	B-Device
7	I-Device
support	O
for	O
any	O
of	O
the	O
X700	O
-	O
X850	O
products	O
.	O
</s>
<s>
In	O
terms	O
of	O
supported	O
DirectX	B-Application
features	O
,	O
R420	B-Device
(	O
codenamed	O
Loki	O
)	O
was	O
very	O
similar	O
to	O
the	O
R300	B-Device
.	O
</s>
<s>
R420	B-Device
basically	O
takes	O
a	O
"	O
wider	O
is	O
better	O
"	O
approach	O
to	O
the	O
previous	O
architecture	O
,	O
with	O
some	O
small	O
tweaks	O
thrown	O
in	O
to	O
enhance	O
it	O
in	O
various	O
ways	O
.	O
</s>
<s>
The	O
chip	O
came	O
equipped	O
with	O
over	O
double	O
the	O
pixel	O
and	O
vertex	O
pushing	O
resources	O
compared	O
to	O
the	O
Radeon	B-Device
9800	I-Device
XT	O
's	O
R360	O
(	O
a	O
minor	O
evolution	O
of	O
the	O
R350	O
)	O
,	O
with	O
16	O
DirectX	B-Application
9.0b	I-Application
pixel	O
pipelines	O
and	O
16	O
ROPs	O
.	O
</s>
<s>
One	O
would	O
not	O
be	O
far	O
off	O
seeing	O
the	O
X800	O
XT	O
basically	O
as	O
a	O
pair	O
of	O
Radeon	B-Device
9800	I-Device
cores	O
connected	O
together	O
and	O
also	O
running	O
with	O
a	O
~	O
30%	O
higher	O
clock	O
speed	O
.	O
</s>
<s>
The	O
R420	B-Device
design	O
was	O
a	O
4	O
"	O
quad	O
"	O
arrangement	O
(	O
4	O
pipelines	O
per	O
quad	O
.	O
)	O
</s>
<s>
This	O
is	O
how	O
the	O
R300-series	O
chips	O
performed	O
their	O
tasks	O
as	O
well	O
,	O
but	O
R420	B-Device
refined	O
this	O
by	O
allowing	O
programmable	O
tile	O
sizes	O
in	O
order	O
to	O
control	O
work	O
flow	O
on	O
a	O
finer	O
level	O
of	O
granularity	O
.	O
</s>
<s>
This	O
changed	O
the	O
ratio	O
of	O
pixel/vertex	O
shaders	O
from	O
2:1	O
(	O
on	O
R300	B-Device
)	O
to	O
8:3	O
,	O
showing	O
that	O
ATI	O
believed	O
the	O
workload	O
in	O
games	O
as	O
of	O
2004	O
and	O
onward	O
to	O
be	O
more	O
pixel	O
shader	O
and	O
texturing	O
oriented	O
than	O
geometry	O
based	O
.	O
</s>
<s>
Normal	B-Application
and	O
parallax	O
mapping	O
were	O
replacing	O
sheer	O
geometric	O
complexity	O
for	O
model	O
detail	O
,	O
so	O
undoubtedly	O
that	O
was	O
part	O
of	O
the	O
reasoning	O
.	O
</s>
<s>
As	O
such	O
,	O
this	O
chip	O
was	O
obviously	O
designed	O
for	O
a	O
heavier	O
geometry	O
load	O
than	O
texturing	O
,	O
perhaps	O
being	O
tailored	O
for	O
a	O
role	O
as	O
a	O
FireGL	B-Application
chip	O
.	O
</s>
<s>
With	O
R420	B-Device
's	O
and	O
RV410	O
's	O
6	O
vertex	O
shaders	O
combined	O
with	O
higher	O
clock	O
speeds	O
than	O
the	O
previous	O
generation	O
,	O
ATI	O
was	O
able	O
to	O
more	O
than	O
double	O
the	O
geometry	O
processing	O
capability	O
of	O
9800XT	O
.	O
</s>
<s>
Although	O
the	O
R420-based	O
chips	O
are	O
fundamentally	O
similar	O
to	O
R300-based	O
cores	O
,	O
ATI	O
did	O
tweak	O
and	O
enhance	O
the	O
pixel	O
shader	O
units	O
for	O
more	O
flexibility	O
.	O
</s>
<s>
ATI	O
revealed	O
Temporal	B-Algorithm
Anti-Aliasing	I-Algorithm
,	O
a	O
new	O
anti-aliasing	O
technology	O
their	O
chips	O
were	O
capable	O
of	O
.	O
</s>
<s>
By	O
taking	O
advantage	O
of	O
the	O
frame-to-eye	O
effects	O
of	O
a	O
framerate	O
higher	O
than	O
60	O
frame/s	O
,	O
the	O
GPU	B-Architecture
is	O
able	O
to	O
better	O
smooth	O
aliased	O
edges	O
by	O
rotating	O
the	O
anti-aliasing	O
sampling	O
pattern	O
between	O
frames	O
.	O
</s>
<s>
Unfortunately	O
,	O
it	O
required	O
the	O
system	O
to	O
be	O
able	O
to	O
maintain	O
at	O
least	O
60	O
frame/s	O
or	O
temporal	B-Algorithm
anti-aliasing	I-Algorithm
would	O
cause	O
a	O
noticeable	O
flickering	O
,	O
because	O
the	O
user	O
would	O
be	O
able	O
to	O
see	O
the	O
alternating	O
AA	O
patterns	O
.	O
</s>
<s>
If	O
the	O
framerate	O
could	O
not	O
be	O
maintained	O
,	O
the	O
driver	O
will	O
disable	O
Temporal	B-Algorithm
AA	I-Algorithm
.	O
</s>
<s>
However	O
,	O
in	O
games	O
which	O
this	O
performance	O
level	O
could	O
be	O
maintained	O
,	O
Temporal	B-Algorithm
AA	I-Algorithm
was	O
a	O
nice	O
addition	O
to	O
ATI	O
's	O
excellent	O
anti-aliasing	O
options	O
.	O
</s>
<s>
Note	O
,	O
ATI	O
's	O
"	O
Temporal	B-Algorithm
AA	I-Algorithm
"	O
was	O
actually	O
a	O
temporal	O
dithering	O
filter	O
for	O
spatial	O
AA	O
,	O
not	O
de	O
facto	O
temporal	B-Algorithm
anti-aliasing	I-Algorithm
(	O
which	O
must	O
involve	O
a	O
controlled	O
blending	O
of	O
the	O
temporal	O
subsamples	O
from	O
consecutive	O
screens	O
)	O
.	O
</s>
<s>
Another	O
notable	O
addition	O
to	O
the	O
core	O
was	O
a	O
new	O
kind	O
of	O
normal	B-Application
map	I-Application
compression	O
,	O
dubbed	O
"	O
3Dc	O
"	O
.	O
</s>
<s>
Similar	O
to	O
how	O
texture	O
compression	O
had	O
been	O
part	O
of	O
the	O
Direct3D	B-Application
specification	O
for	O
years	O
and	O
was	O
used	O
for	O
compressing	O
regular	O
textures	O
,	O
normal	B-Application
map	I-Application
compression	O
compacted	O
this	O
new	O
type	O
of	O
surface	O
detail	O
layer	O
.	O
</s>
<s>
Because	O
DirectX	B-Application
Texture	O
Compression	O
(	O
DXTC	O
)	O
was	O
block-based	O
and	O
not	O
designed	O
for	O
a	O
normal	B-Application
map	I-Application
's	O
different	O
data	O
properties	O
,	O
a	O
new	O
compression	O
method	O
was	O
needed	O
to	O
prevent	O
loss	O
of	O
detail	O
and	O
other	O
artifacting	O
.	O
</s>
<s>
Software	O
making	O
heavy	O
use	O
of	O
normal	B-Application
mapping	I-Application
could	O
gain	O
a	O
significant	O
speed	O
boost	O
from	O
the	O
savings	O
in	O
fillrate	O
and	O
bandwidth	O
by	O
using	O
3Dc	O
.	O
</s>
<s>
Most	O
of	O
the	O
rest	O
of	O
the	O
GPU	B-Architecture
was	O
extremely	O
similar	O
to	O
R300	B-Device
.	O
</s>
<s>
The	O
memory	O
controller	O
and	O
memory	O
bandwidth	O
optimization	O
techniques	O
(	O
HyperZ	B-Device
)	O
were	O
identical	O
.	O
</s>
<s>
R420	B-Device
was	O
actually	O
a	O
secondary	O
4th	O
generation	O
project	O
for	O
ATI	O
,	O
with	O
the	O
original	O
R400	O
plan	O
,	O
internally	O
codenamed	O
"	O
Crayola	O
"	O
,	O
being	O
scrapped	O
.	O
</s>
<s>
R400	O
would	O
have	O
been	O
more	O
feature-complete	O
,	O
with	O
unified	O
Shader	B-Language
Model	I-Language
3	I-Language
shader	O
support	O
among	O
other	O
enhancements	O
,	O
but	O
it	O
is	O
believed	O
that	O
ATI	O
deemed	O
R400	O
unnecessarily	O
complex	O
for	O
the	O
applications	O
that	O
would	O
be	O
available	O
,	O
and	O
potentially	O
risky	O
to	O
develop	O
on	O
the	O
available	O
semiconductor	O
manufacturing	O
processes	O
of	O
the	O
time	O
.	O
</s>
<s>
The	O
R400	O
architecture	O
was	O
thus	O
implemented	O
only	O
in	O
the	O
Xenos	B-Device
chip	O
used	O
in	O
the	B-Operating_System
Xbox	I-Operating_System
360	I-Operating_System
video	O
game	O
console	O
,	O
and	O
became	O
the	O
base	O
for	O
the	O
Qualcomm	B-General_Concept
Adreno	I-General_Concept
200	O
mobile	O
GPU	B-Architecture
,	O
initially	O
called	O
the	O
AMD	O
Z430	O
.	O
</s>
<s>
In	O
the	O
Radeon	O
line	O
,	O
the	O
support	O
for	O
Direct3D	B-Application
9.0c	O
features	O
moved	O
to	O
the	O
subsequent	O
generation	O
based	O
on	O
the	O
R500	B-Device
architecture	O
,	O
while	O
the	O
4th	O
generation	O
was	O
served	O
with	O
the	O
R300-derived	O
R420	B-Device
.	O
</s>
<s>
The	O
earliest	O
Radeon	O
X800	O
series	O
cards	O
were	O
based	O
on	O
the	O
R420	B-Device
core	O
.	O
</s>
<s>
Essentially	O
,	O
the	O
X800	O
Pro	O
is	O
built	O
on	O
semi-defective	O
R420	B-Device
cores	O
.	O
</s>
<s>
R430	O
was	O
unable	O
to	O
reach	O
high	O
clock	O
speeds	O
,	O
being	O
mainly	O
designed	O
to	O
reduce	O
the	O
cost	O
per	O
GPU	B-Architecture
,	O
and	O
so	O
a	O
new	O
top-of-the-line	O
core	O
was	O
still	O
needed	O
.	O
</s>
<s>
The	O
new	O
high-end	O
R4x0-generation	O
arrived	O
with	O
the	O
X850	O
series	O
,	O
equipped	O
with	O
various	O
core	O
tweaks	O
for	O
slightly	O
higher	O
performance	O
than	O
the	O
"	O
R420	B-Device
"	O
-based	O
X800	O
series	O
.	O
</s>
<s>
The	O
"	O
R480	O
"	O
-based	O
X850	O
line	O
was	O
available	O
in	O
3	O
forms	O
:	O
the	O
X850	O
Pro	O
,	O
the	O
X850	O
XT	O
,	O
and	O
the	O
X850	O
XT	O
Platinum	O
Edition	O
,	O
and	O
was	O
built	O
on	O
the	O
reliable	O
high-performance	O
130nm	O
Low-K	B-Algorithm
process	O
.	O
</s>
