|
On May 12th we
showcased our review of NVIDIA's follow-on product to the
NV30,
the GeForce FX 5900 Ultra, based on NVIDIA's NV35
core. This card from the Santa Clara based design
team, is a high end "enthusiast" offering targeted at
delivering the best performance and image quality,
regardless of cost. As such, NVIDIA outfitted their
new flagship with 256MB of DDR Memory at 850MHz. Of
course the Engineering and Marketing teams at ATi had full
knowledge of NVIDIA's impending launch at the time and
began readying their 256MB variant of the Radeon 9800 Pro.
However, as we reported to you back in March, in
our Radeon 9800 Pro launch article and
then again this month in our Radeon 9800 Pro 256MB piece,
ATi was going the way of DDR2 at 700MHz with this new
card, slightly faster and with better latency
characteristics than the 128MB variant, which comes
clocked at 680MHz memory.
At the time we
were testing the new GFFX 5900 Ultra, we didn't get the
new Radeon 9800 Pro 256 into our lab in time to include it
in our battery of benchmarks. Deadlines were looming
and although it would have been possible in theory, to be
able to include this new entrant from ATi, sleep
deprivation took hold of our team and a last minute
decision was made to uphold our "quality" in reporting to
you, rather than quantity. So here we are today,
with a couple more hours of beauty-rest in our overworked
and underpaid bodies and a head to head match up of THE
fastest 3D Accelerators on the market today. We're
going to keep the chatter and technical analysis to a
minimum in this showcase. If you would like a
refresh on where these base architectures came from,
please click the links above. However, we'll provide
you a quick synopsis of the specifics for both cards as
well.
|
|
|
Specifications & Features of The 256MB ATi
Radeon 9800 Pro and The GeForce FX 5900 Ultra |
|
They're big, bad and hella fast |
|
|
Radeon
9800 Pro 256MB
RADEON 9800 Pro 256MB
MEMORY CONFIGURATION
MB
of DDR 2 SDRAM - 350MHz DDR (Effective 700MHz)
3D GRAPHICS FEATURES
- Eight parallel
rendering pipelines process up to 3.04 billion
pixels per second
Four parallel
geometry engines process up to 380 million
transformed and lit polygons per second
High precision
10-bit per channel frame buffer support
256-bit DDR
memory interface
AGP 8X support
SMARTSHADER? 2.1
- Full support
for Microsoft® DirectX® 9.0 programmable pixel and
vertex shaders in hardware
- 2.0 Pixel
Shaders support up to 16 textures per rendering pass
- 2.0 Vertex
Shaders support vertex programs with an unlimited
number of instructions and flow control
- 128-bit per
pixel floating point color formats
- Multiple
Render Target (MRT) support
- Shadow volume
rendering acceleration
- Complete
feature set also supported in OpenGL via extensions
SMOOTHVISION? 2.1
-
State-of-the-art full-scene anti-aliasing
- New technology
processes up to 18.2 billion anti-aliased samples
per second for unprecedented performance
- Supports 2x,
4x, and 6x modes with programmable sample patterns
- Advanced
anisotropic filtering
- Supports up to
16 bilinear samples (in performance mode) or
trilinear samples (in quality mode) per pixel
- 2x/4x/6x full
scene anti-aliasing modes
- Adaptive
algorithm with programmable sample patterns
- 2x/4x/8x/16x
anisotropic filtering modes
- Adaptive
algorithm with bilinear (performance) and trilinear
(quality) options
-
Bandwidth-saving algorithm enables this feature with
minimal performance cost
HYPER Z? III+
- Hierarchical
Z-Buffer and Early Z Test reduce overdraw by
detecting and discarding hidden pixels
- Lossless
Z-Buffer Compression and Fast Z-Buffer Clear reduce
memory bandwidth consumption by over 50%
- Fast Z-Buffer
Clear
- 8.8 : 1 Compression Ratio
- Optimized Z-Cache for enhanced performance of
shadow volumes
|
GeForce FX 5900 Ultra
GEFORCE FX 5900 ULTRA
- .13u
Manufacturing Process
- 256-Bit GPU -
450MHz Clock Speed
- Flip-Chip BGA
Package with copper interconnects
- Up To 8 Pixels Per
Clock Processing
- 1 TMU Per Pipe (16
Textures per unit)
- 2 x 400MHz Internal
RAMDACs
- 256-bit Memory
Architecture
MEMORY
-
850MHz DDR
-
256-Bit Bus Width
-
128MB &
256MB Memory Capacity
-
3rd.
Generation Lightspeed Memory Architecture
-
Effective bandwidth - 27.2GB/s actual @ 850MHz
-
256MB
of DDR/DDR2
3D GRAPHICS FEATURES
- CineFX 2.0 for
Cinematic Special Effects
- "UltraShadow"
Hardware Shadow Acceration
- 2x floating point
pixel shader performance of NV30
- 256MB High Speed
Frame Buffer
- AGP 4X/8x
- DVI + VGA + TV /
VIVO
- Full DirectX 9.0 &
OpenGL Support
CINEFX 2.0
- 64-Bit
Floating-Point Color
- 128-Bit
Floating-Point Color
- Long Program length
for Pixel and Vertex Shading
- Unified Vertex and
Pixel Shading instruction set
- Unified Driver
Architecture
- nView 2.0 -
Multi-Display Technology
- Digital Vibrance
Control 3.0
- 2nd Generation
compression & caching
-
Intellisample HCT
Next
Generation Antialiasing, Anisotropic Filtering and
Compression
-
Hardware Acceleration for Shadows
|
|
|
|
HotHardware's Test Setup |
|
Canterwood and the fastest P4
|
|
Pentium 4
Processors at 3GHz - 800MHz System Bus
Motherboard and RAM Config
Abit IC7-G "Canterwood" Motherboard
512MB of Kingston HyperX PC3500 CAS 2 RAM
CAS Timings were 2-2-2-5
NVIDIA GeForce FX 5900 Ultra 256MB
NVIDIA GeForce FX 5800 Ultra 128MB
ATi Radeon 9800 Pro 256MB
Other
Hardware and Software:
Seagate Barracuda V SATA 120GB HD
Windows XP Professional w/ SP1
NVIDIA Detonator FX Drivers Version 44.03
ATi Catalyst 3.4 Drivers
Intel Release Chipset Driver v5.00.1012
Intel Applications Accelerator RAID Edition v3.0.0.229
|
|
|
3DMark 2001 SE Benchmarks |
|
Direct X 8 Performance |
|
In our first
series of tests, we ran Futuremark's 3DMark 2001 SE and
3DMark 2003. Frankly we're a little uncertain with respect
to validity of this suite of benchmarks, as rumors and
accusations fly in the media of NVIDIA's alleged
"optimizations" (or cheating depending on who you talk to)
on both 2001 and 2003 benchmarks. We have yet to
prove out some of these claims ourselves in our labs, so
for now we'll include these scores as a component of our
total performance metric between the two cards.
These scores are only a piece of the complete picture and
since the tests are "synthetic" in a general sense, they
only correlate loosely with respect to real world gaming
performance.
The GFFX5900
Ultra and the 256MB R9800 Pro are neck and neck here, when
you look at the standard scores, without Anti-Aliasing in
the mix. Actually, the R9800 card has a more
significant lead in the default benchmark, until you
invoke 4X AA. Then the GFFX5900 Ultra pulls ahead by
a healthy margin, with its higher over all memory
bandwidth most likely giving it the edge.
Next Up - 3DMark 2003
and Serious Sam SE
|
