<s>
The	O
RIVA	B-Device
TNT	I-Device
,	O
codenamed	O
NV4	B-Device
,	O
is	O
a	O
2D	O
,	O
video	O
,	O
and	O
3D	O
graphics	O
accelerator	O
chip	O
for	O
PCs	O
that	O
was	O
developed	O
by	O
Nvidia	O
.	O
</s>
<s>
It	O
was	O
released	O
in	O
March	O
1998	O
and	O
cemented	O
Nvidia	O
's	O
reputation	O
as	O
a	O
worthy	O
rival	O
within	O
the	O
developing	O
consumer	O
3D	O
graphics	B-Device
adapter	I-Device
industry	O
.	O
</s>
<s>
The	O
first	O
RIVA	B-Device
TNT	I-Device
based	O
card	O
was	O
released	O
on	O
June	O
15	O
,	O
1998	O
,	O
by	O
STB	O
Systems	O
:	O
Velocity	O
4400	O
.	O
</s>
<s>
The	O
TNT	O
was	O
designed	O
as	O
a	O
follow	O
up	O
to	O
the	O
RIVA	B-Device
128	I-Device
and	O
a	O
response	O
to	O
3Dfx	O
's	O
introduction	O
of	O
the	O
Voodoo2	O
.	O
</s>
<s>
Unlike	O
the	O
Voodoo2	O
(	O
but	O
like	O
the	O
slower	O
Matrox	B-General_Concept
G200	I-General_Concept
)	O
it	O
also	O
added	O
support	O
for	O
a	O
32-bit	O
(	O
truecolor	O
)	O
pixel	O
format	O
,	O
24-bit	O
Z-buffer	O
in	O
3D	O
mode	O
,	O
an	O
8-bit	O
stencil	O
buffer	O
and	O
support	O
for	O
1024×1024	O
pixel	O
textures	O
.	O
</s>
<s>
Improved	O
mipmapping	B-Data_Structure
and	O
texture	O
filtering	O
techniques	O
,	O
including	O
newly	O
added	O
support	O
for	O
trilinear	O
filtering	O
,	O
dramatically	O
improved	O
quality	O
compared	O
to	O
the	O
TNT	O
's	O
predecessor	O
.	O
</s>
<s>
Like	O
RIVA	B-Device
128	I-Device
,	O
RIVA	B-Device
TNT	I-Device
is	O
a	O
single	O
chip	O
solution	O
.	O
</s>
<s>
Originally	O
planned	O
specifications	O
should	O
have	O
placed	O
the	O
card	O
ahead	O
of	O
Voodoo2	O
in	O
theoretical	O
performance	O
for	O
Direct3D	B-Application
applications	O
,	O
but	O
at	O
90MHz	O
it	O
did	O
not	O
quite	O
match	O
the	O
Voodoo2	O
.	O
</s>
<s>
At	O
the	O
time	O
,	O
most	O
games	O
supported	O
3dfx	O
's	O
proprietary	O
Glide	B-Operating_System
API	I-Operating_System
which	O
gave	O
the	O
Voodoo2	O
a	O
large	O
advantage	O
in	O
speed	O
and	O
image	O
quality	O
,	O
and	O
some	O
games	O
only	O
used	O
the	O
Glide	B-Operating_System
API	I-Operating_System
for	O
3D	O
acceleration	O
,	O
leaving	O
TNT	O
users	O
no	O
better	O
off	O
than	O
people	O
who	O
did	O
n't	O
have	O
a	O
3D	O
accelerator	O
.	O
</s>
<s>
Even	O
in	O
"	O
OpenGL	B-Application
only	O
"	O
comparisons	O
such	O
as	O
the	O
case	O
in	O
Quake	B-Application
2	I-Application
,	O
the	O
Voodoo2	O
had	O
the	O
upper	O
hand	O
as	O
a	O
custom	O
"	O
MiniGL	O
"	O
driver	O
was	O
made	O
specifically	O
for	O
3dfx	O
cards	O
to	O
run	O
the	O
game	O
(	O
and	O
most	O
other	O
OpenGL	B-Application
games	O
at	O
the	O
time	O
)	O
.	O
</s>
<s>
The	O
3dfx	O
MiniGL	O
driver	O
was	O
not	O
a	O
fully	O
featured	O
OpenGL	B-Application
driver	O
,	O
but	O
a	O
wrapper	O
that	O
mapped	O
certain	O
OpenGL	B-Application
functions	O
to	O
their	O
equivalents	O
in	O
Glide	B-Operating_System
,	O
and	O
was	O
able	O
to	O
attain	O
a	O
speed	O
advantage	O
because	O
of	O
that	O
.	O
</s>
<s>
Later	O
on	O
when	O
fully	O
featured	O
OpenGL	B-Application
drivers	O
were	O
made	O
for	O
the	O
3dfx	O
line	O
of	O
cards	O
,	O
it	O
was	O
noticed	O
that	O
it	O
was	O
much	O
slower	O
when	O
compared	O
to	O
its	O
cut	O
down	O
MiniGL	O
brother	O
.	O
</s>
<s>
Again	O
,	O
like	O
with	O
the	O
RIVA	B-Device
128	I-Device
,	O
the	O
lack	O
of	O
Glide	B-Operating_System
API	I-Operating_System
support	O
hindered	O
Nvidia	O
's	O
opportunities	O
for	O
market	O
share	O
growth	O
.	O
</s>
<s>
Glide	B-Operating_System
was	O
still	O
considered	O
the	O
best	O
3D	O
gaming	O
API	O
available	O
by	O
both	O
gamers	O
and	O
developers	O
.	O
</s>
<s>
However	O
,	O
TNT	O
gained	O
Nvidia	O
much	O
attention	O
and	O
paved	O
the	O
way	O
for	O
the	O
refreshed	O
version	O
called	O
the	O
RIVA	B-Device
TNT2	I-Device
.	O
</s>
<s>
This	O
board	O
used	O
the	O
same	O
TNT	O
chip	O
but	O
lowered	O
its	O
clock	O
speed	O
and	O
halved	O
both	O
memory	O
data	B-General_Concept
bus	I-General_Concept
width	O
(	O
to	O
64-bit	O
)	O
and	O
memory	O
size	O
(	O
to	O
16	O
MiB	O
)	O
.	O
</s>
<s>
By	O
doing	O
this	O
,	O
Nvidia	O
was	O
able	O
to	O
still	O
sell	O
TNT	O
chips	O
that	O
could	O
n't	O
reach	O
the	O
TNT	O
's	O
specified	O
clock	O
speeds	O
,	O
a	O
practice	O
known	O
as	O
binning	B-Architecture
,	O
and	O
cut	O
board	O
costs	O
significantly	O
by	O
using	O
a	O
narrower	O
bus	O
and	O
less	O
RAM	O
.	O
</s>
<s>
At	O
the	O
time	O
,	O
Quake2	B-Application
was	O
the	O
benchmark	O
for	O
performance	O
and	O
the	O
3dfx-made	O
Voodoo2	O
enjoyed	O
a	O
large	O
performance	O
difference	O
over	O
the	O
TNT	O
because	O
it	O
had	O
3DNow	O
!	O
</s>
<s>
optimizations	O
and	O
the	O
TNT	O
's	O
Quake2	B-Application
performance	O
jumped	O
30%	O
.	O
</s>
<s>
In	O
fact	O
,	O
all	O
OpenGL	B-Application
and	O
DirectX	O
applications	O
benefited	O
from	O
such	O
optimizations	O
.	O
</s>
