The 57 million
transistors encapsulated in the .15 micron GPU give the GeForce
3 the capabilities of the new nFiniteFX engine,
the Lightspeed memory architecture and HRAA
(Quincunx) method. We've included some
screenshots from technology demos provided by
nVidia, highlighting these new features. (The captions were only slightly edited)
The nFiniteFX engine
is the name nVidia gives to the true hardware
support for DirectX 8's vertex and pixel shaders.
The power of these tools has been covered in great
depth on a plethora of other sites, so we won't go
too in depth here. Though these shaders can
be emulated using CPU power, the GF3 handles them
in hardware. These capabilities make the T&L
unit of the GF3
much more powerful than the hardwired T&L units
found on the GF2 and Radeon. The promises we've
been hearing from developers about
super-high poly count models may soon be
fulfilled.
Zoltar:
Zoltar is a fully animated fortune teller.
GeForce3's nFiniteFX engine enables his facial
movements and intricate skin definition.
|
The GeForce 3's
Lightspeed memory architecture is perhaps as
significant a step forward as the nFiniteFX
engine. With the GPU clocked at 200MHz, RAM
clocked at 230MHz (460MHz effectively) and the same
amount of pixel pipes and texture units as the
GeForce 2, some would expect the GeForce 3 to be
slower than a GF2 Ultra...but they'd be wrong.
Although the theoretical peak fillrate is lower
then the 250MHz GPU on a GeForce 2 Ultra, the
increased memory bandwidth realized with the
Lightspeed memory architecture offers performance
as fast or in most cases faster than the GF2
Ultra, even in games that don't take advantage of
DirectX 8. The GeForce 3 can also apply four
textures in a single pass as opposed to two with a
GeForce 2. Even though the GeForce 3 still has
four pixel pipelines, and eight texture units (the
same as the GF2) it can assign two of the
pipelines up to four texture units (leaving the
others idle). In games that blend over three
textures together this results in less reading and
writing to memory, which should yield a
significant performance increase. Not only
does the GF3 texture more efficiently, but it
makes better use of
available memory bandwidth
because of the Lightspeed memory architecture's
Z-Compression, Z-Occlusion Culling and Crossbar
memory controller.
The Z-Compression and
Z-Occlusion Culling are similar the ATi's HyperZ
found in the Radeon. Z-Buffer data is
compressed at a ratio of up to 4:1, and many
undrawn pixels are never rendered alleviating
memory bandwidth needs. The Crossbar memory
controller is a very powerful addition as well.
Instead of using a single 128-Bit memory
interface, the GF3 has four 32-Bit controllers
that each have access to all available memory.
The memory still interfaces at an effective
128-Bits per cycle, but the four controllers can
access data from different places, so much less
bandwidth goes to waste.
Chameleon: The chameleon's
changing skin illustrates just a few of the
effects possible with GeForce3's nFiniteFX
engine.
|
The next advance is
the new Quincunx FSAA method. Unlike the
super-sampling method of the GeForce 2, the
GeForce 3 uses a multi-sampling method that
requires much less fillrate. Instead of
drawing a screen at a much higher resolution and
rendering it down like super-sampling, the new
Quincunx method takes a color sample from four
surrounding pixels and an algorithm calculates the
color value of the final pixel. The end
result is similar in quality to 4X AA, but with
much less of a performance penalty.
Blobby
Dino: This simple technical
demonstration gives a quick look at the
geometry deformation abilities of the Vertex
Shaders in GeForce3's nFiniteFX engine.
|
The Whole
Enchilada: This "Floating Patch" demo
combines the power of GeForce3's Vertex
Shaders, Pixel Shaders, and surface
evaluators.
|
Also migrating to the
GeForce 3 is nVidia's Digital Vibrance feature,
previously found only on MX based cards.
Digital Vibrance allows a user to
adjust color saturation and vibrance to improve
overall image quality. All of us at HotHardware
are fans of this feature, and are pleased it found
it's way into the GeForce 3, the effect it has on
2D image quality is dramatic.
Because our GV-GF3000
was a pre-production sample, we'll keep this next
section very brief...
|
Quality
and Installation Of The Gigabyte
GV-GF3000 |
Short
and Sweet! |
|
The
Gigabyte GV-GF3000 we received was equipped with a
simple heatsink / fan combo. We were not
very impressed with this cooler, but it's going to
be replaced in the final revision so we won't hold
anything against Gigabyte for now.
Our board was not
equipped with TV-Out, but there was a DVI
connector to appease digital flat panel owners.
The GV-GF3000
installed without incident. We did not have
any problems with the card itself, or with
Gigabyte's driver installation.
The CD included with
the GV-GF3000 contained multiple versions of
nVidia's 7.xx reference drivers. However,
our testing was done using the "beta" 11.01
reference drivers.
To the
Benchmarks...
|