| 3DMark 99 (1024x768x32)
At 32-bit colour, the Permedia3's
performance improved quite a bit. Now, after being
overclocked, it was even able to match the standard
RIVA TNT2. Still, the overclocked RIVA TNT2
comfortably beat the overclocked Permedia3 by 29%.
|
The rasterizer score
is calculated from the Texture Rendering Speed
tests and the Fill Rate tests and shows
theoretical 3D performance. This time, the
overclocked Permedia3 managed to beat the standard
RIVA TNT2 by 6% although it's slower than
the overclocked RIVA TNT2 by 16%. It's
evident that the Permedia3 performs comparatively
better in 32-bit colour than it does in 16-bit
colour.
|
This benchmark tests the
single-texturing performance of the card. The
Permedia3 scored about 17.7% lower than the
RIVA TNT2. When overclocked though, it matched the standard RIVA TNT2's
performance but was
still 21.3% slower than the overclocked
RIVA TNT2. Noting the Permedia3's comparatively lacklustre performance in 16-bit colour, one can
only deduce that 3Dlabs had optimized the
Permedia3 for 32-bit
operations.
|
Although the Permedia3 would
perform quite well here in 32-bit colour, its
slower memory apparently has dragged down its
performance at this higher screen resolution. Now,
it's 19.7% slower than the RIVA TNT2.
Overclocking improved its performance, bringing it
almost up to par with the standard RIVA TNT2.
|
This benchmark tests the
single-texturing capability of the card. The
Permedia3 scored quite well here but it's probably
because the RIVA TNT2 isn't so well tuned for
32-bit colour operations. Still, it's interesting
to note that while the overclocked Permedia3
theoretically has a 3% advantage over the
standard RIVA TNT2 in fillrate, the RIVA TNT2 was
still 6.4% faster than it.
|
Surprising, the Permedia3 failed to
match the RIVA TNT2, as we have seen when the
colour resolution was 16-bit. Instead, its
fillrate was only 83.8% of the standard
RIVA TNT2.
After overclocking, the
Permedia3 managed to overtake the
standard RIVA TNT2 but it
was still way slower than the overclocked RIVA
TNT2 - about 14% slower.
|
Because the individual textures are
double the size of the 16-bit textures in the
16-bit colour test, the Permedia3 showed excellent
performance here. It came quite close to the
performance of the RIVA TNT2. In fact, it was only
7% slower than the RIVA TNT2.
|
When the texture size increased to
4MB, the small performance gap closed even more.
In fact, both cards can be said to perform equally
well with 4MB textures in 32-bit colour.
|
With 8MB of textures, the Permedia3
finally surpassed the RIVA TNT2 in texture
rendering speed. Now, the Permedia3's faster than
the RIVA TNT2 by 7-9%. The Permedia3 could
possibly render the textures much faster but for
the hindrance of its low memory bandwidth.
|
Increase the texture size further
to 16MB and the Permedia3 pulls even further away from
the RIVA TNT2. The Permedia3's Virtual Texture
engine now gives it a large 25-27% advantage over
the RIVA TNT2.
|
However, when it came to 32MB
textures, the Virtual Texture completely failed
the Permedia3. Even worse than it was with 16-bit
textures, the Permedia3 was only able to
render the textures at less than a quarter
of the
RIVA TNT2's speed. It would appear that such a
massive load of textures has completely
overwhelmed 3Dlab's much vaunted Virtual Texture
engine.
|
Surprisingly, the Permedia3
performed poorly in this test. After all, in the
800x600x32 resolution, it matched the RIVA TNT2's
performance. Here, it was slower than the RIVA
TNT2 by 26%. It must be the Permedia3's low
memory bandwidth at work again.
|
Again, the Permedia3 performed
poorly. The performance gap actually widened to 39%!
Even when overclocked, the Permedia3 was slower
than the RIVA TNT2 by 27%.
|
However, when it came to the
single-pass bump mapping test, the Permedia3
bounded back and closed the performance gap. In
fact, the overclocked Permedia3 managed to beat
the standard RIVA TNT2 by 3.5%.
|
The Permedia3 apparently can
perform point sample filtering about 22% faster
than it can do bilinear filtering. The RIVA TNT2,
on the other hand, only gained an extra percent of
performance with the simpler point sample
filtering mode.
|
While the RIVA TNT2 obviously
received a slight penalty using the trilinear
filtering mode (~3.5%), the Permedia3 surprisingly gained
a small performance benefit from using trilinear
filtering instead of bilinear filtering! I guess
3Dlabs must have optimized the Permedia3 for the
higher quality trilinear filtering.
|
Again, there's no support for
anisotropic filtering... so it's not surprising to
see that the Permedia3 got another big N/A!
|
|
|
|
