FAQ News Search Archive Forum IRC nZone Articles Tweaks Technology Files Prices Folding
Need Memory?
Popular Articles

NVIDIA GeForce4 Ti 4600 Preview
By: Mike Chambers - February 10, 2002

Performance Testing Configuration

The following is a list of the hardware and software used for performance testing.

  • AMD Athlon XP 1800+ @ 1.53GHz
  • NVIDIA Reference Motherboard (nForce Chipset)
  • 256MB Corsair PC2400 DDR RAM
  • NVIDIA Reference GeForce4 Ti 4600 (300MHz/650MHz)
  • NVIDIA Reference GeForce3 Ti 500 (240MHz/500MHz)
  • Detonator XP Beta Driver Version 27.30
  • 32-Bit Color With The Exception of 2048x1536
  • Sound Disabled / Vsync Disabled / 75Hz Refresh Rate (60Hz @ 2048x1536)
  • Windows XP Professional / DirectX 8.1

Benchmark results are reported in average frames per second unless otherwise stated.

OpenGL Performance - Quake 3

OpenGL game performance begins with version 1.30 of Quake 3 using the Demo4 map. The first series of benchmark results were carried out with the default high quality setting along with maximum quality which consisted of increasing geometry and texture detail. All results thereafter are based on the maximum quality setting.

Quake 3 Test Settings

Setting Description
Standard High Quality
Maximum Quality Include High Geometry and Maximum Texture Detail
Increased Geometry r_subdivisions -1
r_lodbias -2
Complex Shadows cg_shadows 2

High Quality / Maximum Quality

Resolution GeForce4 Ti 4600 GeForce3 Ti 500 Percent Change
High Quality
1024x768 Avg: 220 Avg: 208 + 6%
1280x1024 Avg: 196 Avg: 164 + 20%
1600x1200 Avg: 158 Avg: 122 + 30%
2048x1536 Avg: 117 Avg: 84 + 39%
Maximum Quality
1024x768 Avg: 211 Avg: 191 + 10%
1280x1024 Avg: 178 Avg: 139 + 28%
1600x1200 Avg: 137 Avg: 102 + 34%
2048x1536 Avg: 102 Avg: 71 + 44%
* Results for 2048x1536 are in 16-bit color

Additional load is place on both graphics cards by increasing the geometry used by Quake 3. This was accomplished by changing the values of the following variables.

  • Changing r_subdivisions from 4 to -1 results in the use of a greater number of polygons to render curves.
  • Changing r_lodbias from 0 to -2 produces more rounded images in character models and objects.

Note that the r_lodcurverror variable was left at it's default setting of 250 since it's been cheat protected. Here are two comparative screenshots showing the differences in geometry. Since game play in Quake 3 is fast paced, it's unlikely that you'll actually notice a difference.

High Geometry

Increased Geometry

The front and rear arches, as well as the front light in this archway exhibit smoother curves by increasing the level of geometry. Even the arches in the background have better structural definition. Let's see what happens to performance when these settings are enabled.

Increased Geometry

Resolution GeForce4 Ti 4600 GeForce3 Ti 500 Percent Change
Maximum Quality
1024x768 Avg: 198 Avg: 181 + 9%
1280x1024 Avg: 169 Avg: 134 + 26%
1600x1200 Avg: 133 Avg: 99 + 34%
2048x1536 Avg: 99 Avg: 49 + 43%
* Results for 2048x1536 are in 16-bit color

With either of these graphics cards, you might as well enable the increased geometry settings in Quake 3 since there wasn't a significant decrease in performance from the previous maximum quality setting.

To bring on the deathblow, complex shadows were enabled (cg_shadows 2).

Complex Shadows

In lieu of a semi-darkened circle appearing under a player, complex shadows adds a significantly higher quality shadow to player models and objects such as weapons and power-ups.

Complex Shadows

Resolution GeForce4 Ti 4600 GeForce3 Ti 500 Percent Change
Maximum Quality + Increased Geometry
1024x768 Avg: 143 Avg: 108 + 32%
1280x1024 Avg: 98 Avg: 72 + 36%
1600x1200 Avg: 71 Avg: 52 + 37%
2048x1536 Avg: 49 Avg: 35 + 40%
* Results for 2048x1536 are in 16-bit color

As is typical with many modern games, projecting dynamic shadows with just one light source can be a complex task to develop and causes a significant performance hit. Advancements in development of dynamic shadows are ongoing as John Carmack is developing a robust lighting system with the upcoming Doom which NVIDIA's Mark Kilgard discusses in his CEDEC 2001 -- Robust Stencil Shadow Volumes presentation (1.26MB PDF).

As a final test of this set of benchmarks, I purposely underclocked the GeForce3 Ti 4600 to match the 240MHz processor and 500MHz memory speeds of the GeForce3 Ti 500 using NVmax. This test offers a sneak peak at the benefits of the second iteration of the Lightspeed Memory Architecture (LMA II). Maximum quality settings were used for this test.

Benefits of LMA II

Resolution GeForce4 Ti 4600 GeForce3 Ti 500 Percent Change
250MHz/500MHz - Maximum Quality
1024x768 Avg: 198 Avg: 191 + 4%
1280x1024 Avg: 148 Avg: 139 + 6%
1600x1200 Avg: 109 Avg: 102 + 7%
2048x1536 Avg: 80 Avg: 41 + 13%
* Results for 2048x1536 are in 16-bit color

It was surprising to learn that the faster clock speeds of the GeForce4 Ti 4600 had contributed to the majority of the performance increase in the Quake 3 tests. Assuming that the increased performance of the GeForce4 Ti 4600 can be attributed to occlusion culling enhancements, the results make sense as the depth complexity in Quake 3 is far less than that found in the Villagemark benchmark which was used in Anandtech's GeForce4 preview. Their results showed a performance increase of 25% for the GeForce4 Ti 4600 when both cards were clocked at 240MHz/500MHz. It's difficult to pinpoint exactly what enhancements to the LMA II lead to this increase, but the extra performance is certainly welcome.

To wap up this section, both the GeForce4 Ti 4600 and GeForce3 Ti 500 are providing more than adequate performance to enjoy Quake 3 at super-high resolutions. However, the additional performance of the GeForce4 Ti 4600, which averaged 29.3% for all these tests, will be better served in modern games or additional enhancements to graphics quality.

Next Page: OpenGL Anisotropic Performance - Quake 3

Skip To:


Last Updated on March 8, 2002

Copyright 1998-2003. All rights reserved.
Reproduction in any form or medium without written permission of the site's owners is prohibited.

Privacy Policy

FastCounter by bCentral

 Visitors Are Online
Powered by Perlonline.com
Graphics Utilities
Add-In Cards
For Developers