facebook rss twitter

Review: Gainward GeForce FX 5900 Ultra 256MB

by Tarinder Sandhu on 17 August 2003, 00:00 4.0

Tags: Gainward

Quick Link: HEXUS.net/qasp

Add to My Vault: x

Comparative specs

Radeon 9800 Pro Gainward FX 5900 Ultra GeForce FX 5800 Ultra (NV30)
Manufacturing process 0.15 0.13 0.13
GPU Speed 380MHz 300/450 (2D/3D) 300/500 (2D/3D)
Memory speed 680MHz DDR-I 850MHz DDR-I 1000MHz DDR-II
Memory interface 256-bit 256-bit 128-bit
Maximum memory 256MB 256MB 128MB
Memory Bandwidth 21.8GB/s 27.2GB/s 16GB/s
Triangle Throughput 380 MT/s 700 MT/s ??? 350 MT/s
Pixel Fillrate (ST) 3.04 GP/s 1.80 GP/s 2 GP/s
Pixel Fillrate (MT) 3.04 GT/s 3.60 GT/s 4 GT/s
AA Fillrate 18.24 Billion/s 14.4 Billion/s 16 Billion/s
Rendering Pipelines 8 8/4* 8/4*
Textures Per Pipe 1 2 2
Textures Per Pass 16 16 16
Vertex Shader 2.0+ 2.0 2.0+
Vertex Shaders 4 Floating Point Array Floating Point Array
Vertex instructions 65,536 65,536 65,536
Pixel Shader 2.0 (F-buffer) 2.0 + 2x FX 5800U ability 2.0+
Pixel instructions Unlimited 1,024 1,024
Pixel precision 96-bit (4 x 24-bit) 128-bit (4 x 32-bit) 128-bit (4 x 32-bit)
FSAA 6x 8xS 8xS
FSAA Method Multisampling Multisampling / SS Multisampling / SS
Bandwidth saving HyperZ III+ LMA III LMA III
Image enhancement SmoothVision 2.1 Intellisample HCT Intellisample
AGP rates 1x/2x/4x/8x 1x/2x/4x/8x 1x/2x/4x/8x
Connections TV-Out, VGA, DVI TV-Out, VGA, DVI TV-Out, VGA, DVI
Display 2x 400MHz DACs 2x 400MHz DACs 2x 400MHz DACs
Special Features F-Buffer, Stencil shadow buffer UltraShadow DDR-II memory

Built on a 0.13-micron process and boasting faster GPU and memory clocks than the Radeon 9800 Pro, the Gainward FX 5900 Ultra should be faster than the Canadians' card. However, its method of rendering may throw a spanner in the works, especially with games that rely heavily on colour + z-pixel rendering.

* 8/4 - which is it ?

From NVIDIA themselves: The NV30/35 runs at 8 operations per clock for the following scenarios.

z-rendering
stencil operations
textures operations
shader operations

But color + z-pixel rendering is done at 4 operations per clock. This is by far the most common usage in today's games. 8-pipeline running may come into play in the future, though it's a decidedly 4-pipeline approach for today.