| |

Introduction
In an eerie half light a tumble weed
rolls silently across a sparse, featureless landscape.
Vultures circle above waiting for the next set of
weary bones to pick clean while earthbound scavengers
scurry through the baking dust. Oh yes, the world
of the graphics card outcast isn't a good place to
holiday by any stretch of the imagination. Ati know,
they've been there, and now NVIDIA know it too as
they've been looking for somewhere to check in there
for a few weeks now.
Much as I want to defend NVIDIA for
the huge impact they've had on the graphics industry
I'm not going to. They're big boys now and I'm sure
they're perfectly capable of taking the punches being
thrown at them squarely on the chin. Nor am I going
to be rubbing salt into the already festering wounds.
They're a young and savvy company who don't need the
likes of me pointing out their mistakes to them. They've
become all too painfully aware of these mistakes for
themselves and will no doubt learn from them.
Let's therefore summarise the situation
in simple, factual terms with a little opinion thrown
in for good measure. Way back at the dawn of time
NVIDIA launched a legendary card, the GeForce. This
card redefined the computer graphics industry in way
none of could have imagined. It may not have been
the fastest of cards by today's standards but it showed
that moving some of the burden of rendering a complex
3D scene from the CPU to a dedicated GPU (Graphics
Processing Unit) was indeed feasible. Graphics cards
from here on in would never be the same again.
For many a year Ati struggled to introduce
a product that could compete, and yet good though
some of them were, they never quite had what it takes
to dethrone the undisputed graphics champions. That
is until the introduction of a graphics card that
if I'm being brutally honest, deserves far more plaudits
than it has so far received, the Radeon 9700 Pro.
The Radeon 9700 Pro took everything that NVIDIA had
achieved and packaged it in a compact, unspectacular
looking graphics card that set new standards for visual
quality and performance. A card that in many ways
deserves a similar legendary status to that of the
first GeForce.
NVIDIA, being the dynamic company they
are fought back with a card doomed to difficulties
before it ever reached the taping out stage. The move
from 0.15 to 0.13 microns was proving a major headache,
yields were so poor that the card wasn't likely to
become financially viable in any reasonable time frame,
and the gamble on the readiness of DDR2 to power their
beast was at best optimistic. DDR2 was hot, expensive
and far from ready for a role in mainstream graphics
cards. And to cool the toasty DDR2 memory chips NVIDIA
also needed a far more aggressive cooling system than
the market had seen before on a mainstream product,
an incredibly noisy cooling system that went on to
become the butt of many a cruel joke as NVIDIA tried
in vain to defend their decisions and sell the idea
to an increasingly picky public. NVIDIA couldn't understand
it, when they last looked, just before they all got
buried chin-deep in the task of getting the NV30 ready
to roll, noisy PC's were all the rage? People were
splashing out tens of dollars on CPU coolers that
came with dual, high RPM fans that sounded like someone
drilling a hole in a rock. What they hadn't realised
while they'd been so totally absorbed in preparing
their new baby for the big time was that PC users
had grown up. Gone was the need for noisy, costly,
brick sized copper coolers, fans that could slice
carrots and power supplies that sounded like a jet
engine and in came the quest for the silent PC. Heatpipes,
thermally controlled fans and large, passive sinks
had replaced loud and crude brut force cooling and
this just made the troubled NV30 based GeForce FX5800
Ultra stand out like a sore thumb. Further tamed by
its suffocating 128bit memory bus the FX 5800 Ultra
was always going to be, if you'll excuse the term,
peeing against the wind.
Okay, so NVIDIA made a mistake, a misjudgment,
they're not the first company to do it and they won't
be the last. What's happened since then however with
regards 3DMark03 is for you to read about on other
sites. I know too much about what's gone on behind
the scenes to make any fair and unbiased comments
but let's just say the problem isn't restricted to
NVIDIA and that turning the whole event into a witch
hunt is achieving nothing towards the final goal of
offering accurate and repeatable benchmarks that fairly
and accurately reflect a graphics card's true performance
and potential.
Anyway, back to the tale. Realising
that the 5800 Ultra was doomed to failure NVIDIA were
swift to tackle its major weaknesses, those of noisy
cooling, heat creation and a rather limiting 128bit
memory bus, by dropping it like a hot rock and replacing
it with a new product, the FX 5900 Ultra, and it's
this very card I'm looking at today. They could have
redesigned the cooling to make it quieter but chose
not to, this in itself suggests there was more to
dropping the 5800 Ultra than just noise. Talk is that
that the 5800 Ultra was just too complex and expensive
for NVIDIA partners to tool up and start manufacturing
for themselves, then again there's always talk of
something or another. What matter at this stage is
that the FX 5900 Ultra is here and those who shelled
out on a 5800 Ultra, who may well be thinking they
were well and truly stitched up, still have a card
that performs well and is likely to become a collector's
item in time so perhaps it's not as bad as it seems.

|
FX
5900 Ultra |
FX
5800 Ultra |
9800
Pro |
9700
Pro |
| GPU
Die Size |
0.13
|
0.13
|
0.15
|
0.15
|
| Core
Speed |
450
MHz
|
500
MHz
|
380
MHz
|
325
MHz
|
| Memory
Frequency |
425
MHz DDR-I
(850 MHz)
|
500
MHz DDR-II
(1000 MHz)
|
340
MHz DDR-I
(680 MHz)
|
310MHz
DDR-I
(620 MHz)
|
| Memory
Bus |
256-bit
|
128-bit
|
256-bit
|
256-bit
|
| Memory
Bandwidth |
27.2
GB/s
|
16
GB/s
|
21.8
GB/s
|
19.8
GB/s
|
| Pixel
Fill Rate |
3.6
GP/s
|
4
GP/s
|
3.04
GP/s
|
2.5
GP/s
|
| Render
Pipelines |
8
|
4
|
8
|
8
|
| Textures
Per Pipeline |
2
|
2
|
1
|
1
|
| Vertex
Shader |
2.0+
|
2.0+
|
2.0+
|
2
|
| Vertex
Shaders |
Floating
Point Array
|
Floating
Point Array
|
4
|
4
|
| Pixel
Shader |
2.0+
|
2.0+
|
2.0
(F-buffer)
|
2.0
|
| Pixel
Instructions |
1024
|
1024
|
Unlimited
|
64
|
| Anti-Aliasing |
Off
- 8X
|
Off
- 8X
|
Off
- 6X
|
Off
- 6X
|
| AA
Type |
Multisampling
|
Multisampling
|
Multisampling
|
Multisampling
|
| Bandwidth
Saving Techniques |
LMA
III
|
LMA
III
|
HyperZ
III+
|
HyperZ
III
|
| Enhancements |
Intellisample
HCT
|
Intellisample
|
SmoothVision
2.1
|
SmoothVision
2.0
|
| AGP
Rates |
1X
- 8X
|
1X
- 8X
|
1X
- 8X
|
1X
- 8X
|
| RAMDAC |
2
x 400MHz DACs
|
2
x 400MHz DACs
|
2
x 400MHz DACs
|
2
x 400MHz DACs
|
| Connections |
TV-Out,
VGA, DVI
|
TV-Out,
VGA, DVI
|
TV-Out,
VGA, DVI
|
TV-Out,
VGA, DVI
|
The GeForceFX 5900 Ultra now features
an extra 5 million transistors weighing in at a whopping
130 million of them! It has also increased the FX5800's
memory bandwidth from 16 to just over 27GB/sec. On
the downside we see memory frequency dropped from
500MHz to 425MHz (effectively 1GHz to 850MHz) and
the core clock reduced from 500MHz to 450MHz. The
400MHz RAMDACs, 128bit colour precision, 0.13micron
process and full support for DirectX9.0 make it through
intact.
|
|