Posted by Thomas
McGuire on April 25, 2002
Manufacturer: eVGA Product:
MX440 prices / videocard
As you can guess by
the name, this is an updated feature present on the previous
GeForce 3 boards. LMA 1/2 both work to reduce memory
bandwidth requirements, which will increase performance by
reducing redundant processing. The ATI Radeon & PowerVR
Kyro Graphics cards feature similarly themed technologies
– reduce redundant processing, save bandwidth, improve
performance. It does this in several ways:
memory controller. Compared to
the GeForce 4 Ti, the GeForce 4 MX uses 2 Crossbar memory
controllers, as opposed to 4. This provides for a more
efficient way of accessing the frame buffer as compared with
traditional memory controllers. The net effect being that it
optimises bandwidth use. No application support is required
Z-Compression. The Z-Buffer
contains depth information for every single pixel to be
rendered by the Graphics card. Thus using higher resolutions
not only consume more fill rate, they also require a larger
Z-Buffer as more pixels will be rendered. The GeForce 4 does
on-board hardware compression of Z-Buffer data. Most
critically it is loseless, so there will be no precision
loss which could adversely affect the image. According to
NVIDIA this hardware compression will reduce the Z-Buffer
requirements about 25%. Once again, this saves memory
bandwidth. No application support is required for this.
Graphics cards have performed a lot of redundant rendering
by rendering pixels that will inevitably not be visible.
This redundant rendering is known as overdraw. The level of
overdraw can vary from title to title with some having an
overdraw level of 3 or more, while others are much lower.
Reportedly id Software game engines’ are excellent in this
area. Though clearly with games getting more complex the
level of overdraw is not about to decrease of it’s own
accord. While many Developers have added in their own
software methods to reduce overdraw others will need to rely
on that provided in Graphics cards themselves, if any, e.g.
PowerVR Kyro cards use a Tile-based deferred rendering
system which is excellent at eliminating overdraw & are
perhaps the most efficient consumer Graphics cards around.
The first part of
this process is the Depth test - Z-Buffer values are
compared to determine what is visible in a scene, i.e.
pixels at the front of a scene will obscure pixels with a
greater depth, i.e. behind them. For example, if your
viewpoint is looking at a solid wall, then any pixels behind
that wall won't be visible through it. As such any
non-visible scene elements are completely disregarded, only
visible elements get sent on to the Framebuffer.
The second is the
Occlusion Query. This tests whether triangles are hidden
completely behind other triangles. As before, non-visible
triangles are disregarded & play no further part in the
Culling, unlike the other LMA
II features does however require Application support &
will only work with Front to Back order rendering Titles.
Pre-Charge. This minimizes the
latency issues involved with accessing a different area of
memory. No application support is required for this.
self-explanatory this one. This minimizes the amount of
bandwidth & time needed to clear Z-Buffer data which
will also improve performance further. No application
support is required for this.
The Graphics card
comes bundled with the great PowerDVD XP (4.0), albeit
limited to 2 Channel or S/PDIF Output. This is a fairly poor
choice, as it happens the supplied PowerDVD 4.0 doesn’t
support the MPEG decoding features of the GeForce 4 MX. When
trying to play a DVD with Hardware acceleration enabled
you’ll be greeted with the following non-descript error
Thankfully, there is
a patch available to fix this though however take note
because it’s not listed on eVGA’s website and wasn’t
easy to find on Cyberlink’s website either, eventually I
managed to locate it here
This patch is designed for anyone who purchased PowerDVD XP
4.0 online before March 8, 2002 or anyone who received
PowerDVD XP 4.0 bundled with their PC hardware. It provides
support for the most recent devices on the market, such as
ATI TwinView & NVIDIA NV17/NV 25 chips.
The GeForce 4 MX
itself features NVIDIA’s VPE (Video Processing Engine)
which provides Motion Compensation & iDCT (inverse
Discrete Cosine Transform) support for hardware assisted
MPEG decoding, which can greatly reduce CPU usage during DVD
playback & such.
In most instances,
very little are the changes between frames on DVD (MPEG2)
movies. Using predictive coding it is possible to
calculate/measure the motion of moving objects between
frames, though. This data can then be used to predict their
position in future frames. A supported software DVD decoder
is required for this, though the bundled PowerDVD XP will
provide you with this support once patched.
To test out the MPEG
decoding features of the GeForce 4 MX I loaded up the newly
patched PowerDVD XP and ran a 5 minute segment of “The
Fast and the Furious”. Audio Output mode was set to S/PDIF
Output as this would allow the SoundBlaster Audigy’s
Driver to handle the AC-3 decoding, providing 5.1 output.
As you can see there’s not too much difference in CPU
usage either way, though I’d guess this can also be scene
dependent as well. For comparisons sake the 3D Blaster
GeForce 3 Ti 200 which I had previously tested with in Pearl
Harbor showed a much more noticeable reduction in CPU usage
when Hardware Acceleration was enabled.