a fairly uneventful 2003, the graphics marker is suddenly
looking healthier and more exciting than it has in a long
time. With NVIDIA's reputation riding on NV4x and ATi determined
to prove the R300 was no fluke, it was always going to be
a thrilling year for gamers and gadgeteers.
both NVIDIA and ATi heading up the push towards genuine,
real-time cinematic graphics it seems the last few generations
have delivered what they claimed they would but not at sufficiently
fast framerates to make their use practical for anything
but the most basic titles which tended not to support them
days gamers have started adding the words "image quality"
to a vocabulary that used to consist of nothing more than
the words "frame" and "rate", and the
expectations from high powered GPUs and VPUs is at an all-time
high. Anti-aliasing and Anistropic filtering are no longer
pleasant features for those rare occasions they can be applied,
they're now a essential part of the gaming experience and
to many people the idea of running without them is simply
websites worldwide queuing up to sniff out "cheats"
and "unfair optimizations", gamers screaming for
better and faster eye candy and lower prices and partners
begging for a saleable product and a few months to try and
earn an honest crust from them, ATi and NVIDIA were being
pushed all the way to the tape, or should that be all the
way to the tape out?
the dust has finally settled and a few truths seem to have
surfaced. The first is that, on balance, NVIDIA appear to
have won this latest round on points, but even that could
be open to debate as it seems positions at the top are still
in a state of flux. Strangely though, ATi haven't felt the
backlash often felt by the "runner up" in these
battles, perhaps a reflection of the respect they earned
for themselves during NVIDIA's recent public flogging over
NV3x. Perhaps also, and I hope this the case, people are
realising that there will always be a winner and a second
place at least and that doesn't always translate as a superior
and an inferior product. Maybe at last we're begining to
get over our framerate mentality and see the bigger picture.
I'm growing a touch tired of seeing the same equation trotted
out at every possible opportunity. You know the one,
if Card A = 120fps and Card B = 122fps then Card B = Better.
It doesn't work, it never worked and it probably never will
work. That's not to say the faster card can't be the better
card, it just means don't bet your house on it until all
the facts are in.
the subject of today's review is the Radeon X800 Pro courtesy
of SAPPHIRE Tech. Running at 475MHz for the core and 900MHz
for the memory (450MHz base), the biggest difference between
the pro and the XT is that the Pro runs off just 12 of the
available 16 pipelines while the XT uses all 16. How much
will this hurt performance? Let's see.
Here's a quick feature rundown lifted straight off their
POWER CONNECTION REQUIREMENTS
of 'Xtreme performance GDDR3 memory accelerates the
latest cutting edge 3D applications
quad-channel GDDR3 memory interface
'Xtreme parallel pixel pipelines
the AGP 8X and AGP 4x standard, providing a high-speed
link between the graphics board and the rest of the PC
programmable vertex shader pipelines
support for DirectX® 9.0 and the latest OpenGL® functionality
SMARTSHADER™ HD technology allows for support for
Microsoft® DirectX® 9.0 programmable vertex and pixel
shaders in hardware as well as OpenGL® via extensions.
HD technology provides enhanced image quality by removing
jagged edges and bringing out fine texture detail, without
High quality 4:1 Normal Map Compression delivers beautiful
scenes without the performance hit.
VIDEOSHADER™ HD engine uses programmable pixel shaders
to accelerate video processing and provide better-looking
Z™ HD is optimized for performance at high display resolutions,
including widescreen HDTV resolutions.
RADEON™ X800 PRO must be connected to the computer? power
supply to meet its performance and high-speed memory requirements.
Remove the power cable from the hard drive.
In order to supply the ATI Graphics Accelerator card with
the necessary power, we have included the Power Extension
Use the Power Extension Cable to connect the Radeon X800
PRO to the computer? hard Drive power connector.
2. Connect the 4-pin connector of the Power Extension Cable
to the RADEON™ X800 PRO power connector.
3. Connect the Power Extension Cable to the power supply connector.
4. Connect the Power Extension Cable to the Hard Drive connector.
in both FULL Retail and Lite Retail Versions
retail indicates that the package contains the entire bundle
including the Game software bundle and utilities. The Lite
retail versions do not come bundled with the software bundle
and are packaged with the neccessities needed to enjoy the
pinnacle in performance from Sapphire alone.
Card Specifications are subject to change dependent upon
region purchased in and subject to component availability.
Sapphire has enjoyed a commitment to excellence and to its
customers so please do not hesitate to inquire about specific
products at your regional Sapphire office.
not pull any punches here, the R420 isn't a massively different
chip to its predecessor the R300. There are however a few
important additions and refinements so before we move on
let's take a quick look at the more important ones.
of building even the most simple of objects from many tens
of thousands of polygons, the best way to add detail to
the surface of an object is though the clever use of high
resolution textures or manipulation of the available lighting
information. Imagine a low ploygon model, it looks angular
and false. Now drape it with a high quality texture and
suddenly it loses some of its angles and starts to resemlble
what it's meant to be. At this stage though the texture
still looks flat and lacking any depth and real detail.
than pump up the polygon count, a far lest costly way to
increases surface detail is to use high resolution normal
Normal is a piece of data showing the way in which light
interacts with a surface at a given point. The problem is
that models made from too few polygons contain too few normals
to create a properly lit surface. What 3DC does is to create
a high resolution normal map which stires vastly more information
than would normally be generated and then apply that to
the object, so giving the illusion that it contains far
more detail than it really does.
this method uses pixel operations rather than geometry functions,
it tends to be more efficient on consumer graphics cards
which are often optimized for pixel data throughput.
course the benefits from 3Dc are slowly being eroded as
polygon counts and texture quality inreases in modern games
but ultimately it's a feature worth having. 3Dc works on
any two-channel data formats and is not particularly limited
to normal maps.
antialiasing is a simple yet brilliant idea and may well
set the trend for sharing workloads across multiple frames.
The idea is this, apply 4x anialiasing but rather than apply
it all in a single frame, split the load across two adjacent
frames. Brilliant! 4xAA for the same cost as 2x, 8x for
the same cost as 4x and 12x for the same cost as 6x. Well,
brilliant provided the framerate is high enough, a flickering
mess if it's not which is why lower threshold is set via
can't help but ponder the other possibilities here. Could
we see super-high resolution textures applied progressively
across two frames? How about ultra AF where the foreground
of a scene is rendered in frame one and the background in
frame 2. Perhaps my dreams that on-the-fly dynamic IQ levels
could see everything cranked up when a scene slows and we're
more likely to be admiring the scenery and things gradually,
and without intervention, get lowered as things become more
definition gaming with up to 16 pipelines, up to 80 concurrent
shader operations per clock, 6 vertex shaders delivering
in excess of 750million vertices per second, improved Centroid
sample and new Temporal Antialiasing plus FULLSREAM video
deblocking, HDTV support, HYPER Z HD and more, and all in
a single slot solution.
tempting? Let's look at the hardware.