Value Video Card Showdown
By Robert Heron - Posted Nov 06, 2002
The Since we test video cards using the most strenuous settings possible (best visual quality), we were forced to dig around the Windows Registry in order to put the Xabre on (somewhat) equal terms with the rest. SiS's video control panel offers no quality/performance adjustments -- shady indeed. In addition, the Xabre 400 is the only card in this comparison that uses a software solution for its vertex shader support. The DirectX 8 specification allows for this, but our other contenders provide vertex shader support in hardware -- where it should be.Testing DirectX-based game performance
We used the newly released demo of "Unreal Tournament 2003" (UT2K3) as well as 3DMark2001 SE. Epic Games' Unreal Tournament 2003 is a popular first-person shooter that features highly detailed character models and environments that stress CPUs and video cards alike. Although UT2K3 includes an integrated benchmark tool, we created our own maximum-quality configuration file that further increases the difficulty of this test. See comparison chart.
MadOnion's 3DMark2001 SE is a DirectX-based video card benchmark tool that's free to download. 3DMark2001 runs through several challenging 3D scenes and generates a final score based on the number of individual tests completed as well as how quickly they were performed. See comparison chart.
OpenGL testing
We loaded up LucasArts' "Jedi Knight II: Jedi Outcast." Jedi Knight II is based on a modified "Quake III" game engine that incorporates support for hardware shaders. Compared to the original "Quake III," "Jedi Knight II" is far more stressful on a PC. See comparison chart.
Settings and configurations
- All software was configured to use a 32-bit color depth and 32-bit textures. To make the tests a bit tougher, we disabled all texture compression.
- We enabled trilinear filtering and triple buffering when possible.
- Anisotropic filtering (AF) is another computationally intensive form of texture filtering that's gaining in popularity because of the excellent results it produces. We set all of our video cards to use 8x AF except for the Xabre, because we're unsure if it even supports it properly (and we were growing tired of hunting through the Windows Registry). Individual applications can override the AF settings in the video control panel, but we enabled it so applications could take advantage of it when possible.
- Besides maxing out the graphical settings of each program, we also tested how well each card was able to perform with 4x antialiasing.
Test system
- Pentium 4 2.53-GHz processor
- Intel D850EMV2 motherboard
- 512MB of PC800 RDRAM
- Fresh install of Windows XP Pro with SP1
- Latest publicly available video drivers were used with each card
Sound a key factor
Our test system used the integrated sound hardware of the Intel motherboard. This last item is something to keep in mind as you look at our results. Many reviewers test video cards without sound hardware in order to better isolate the video hardware's performance. We're of the opinion that nobody games without sound, so why benchmark with sound disabled? Both sides in this ancient benchmarking issue offer valid points as to why one way is better than the other, and we wanted to be clear about how we perform our testing. Enabling sound lowers performance, but we feel it offers a more realistic picture of how a particular product will perform.Results for the "Unreal Tournament 2003" demo at TechTV's maximum-quality configuration are measured in fps. Higher is better.
|
Video card
|
ATI Radeon 9700 Pro
|
Leadtek WinFast A250 Ultra TD (GeForce4 Ti4600)
|
|
MSI GeForce G4 Ti4200-TD (64MB GeForce4 Ti4200)
|
SiS Xabre 400 (128MB)
|
ATI Radeon 9000 Pro (64MB)
|
|
800x600
|
126.5
|
52.7
|
43.4
|
43.8
|
27.8
|
Failed
|
|
1024x768
|
94.2
|
33.8
|
27.8
|
28.1
|
19.4
|
Failed
|
|
1280x1024
|
63.8
|
21.8
|
17.9
|
18.1
|
12.7
|
Failed
|
Analysis
Our maximum-quality configuration may require some further tweaking as it pushes every one of our value contenders under 50 fps. The failure of the Radeon 9000 Pro to complete this "flyby" benchmark is likely due to a combination of our aggressive detail configuration and a driver issue. We're confirming with the game's publishers that our configuration isn't doing something odd and causing such low results -- that is, low for everything but the Radeon 9700 Pro.
With this extreme example, the additional cost of the 9700 Pro translates into significant performance gains that appear to make it a worthwhile investment. We'll be revisiting this benchmark again after confirming the settings of our test.Here are the final scores for the 3DMark2001 SE benchmark running at 1024x768 at maximum quality. Higher is better.
|
Video card
|
ATI Radeon 9700 Pro
|
Leadtek WinFast A250 Ultra TD (GeForce4 Ti4600)
|
ATI Radeon 9000 Pro (64MB)
|
MSI GeForce G4 Ti4200-TD (64MB GeForce4 Ti4200)
|
Gainward GeForce4 PowerPack! Ultra 650/XP (128MB GeForce4 Ti4200)
|
SiS Xabre 400 (128MB)
|
|
No AA
|
13,307
|
7,686
|
7,071
|
6,437
|
6,387
|
4,496
|
|
4x AA
|
10,075
|
5,027
|
1,984
|
3,953
|
3,910
|
Failed
|
Analysis
With 32-bit color and uncompressed 32-bit textures, all of our contenders were able to complete the 3DMark2001 SE benchmark with antialiasing (AA) disabled. The 128MB Xabre 400 was the only card that was unable to run this test in "pure hardware" mode because of its lack of a hardware vertex shader. Again, despite the significant additional cost of a Radeon 9700 Pro, it manages to produce scores that are at least two times greater than our value video cards. Interestingly, the sub-$100 Radeon 9000 Pro takes first place among our value picks when AA is disabled.Results for OpenGL testing running "Jedi Knight II: Jedi Outcast" at maximum quality are measured in fps. Higher is better.
|
Video card
|
ATI Radeon 9700 Pro
|
Leadtek WinFast A250 Ultra TD (128MB GeForce4 Ti4600)
|
ATI Radeon 9000 Pro (64MB)
|
Gainward GeForce4 PowerPack! Ultra 650/XP (128MB GeForce4 Ti4200)
|
MSI GeForce G4 Ti4200-TD (64MB GeForce4 Ti4200)
|
SiS Xabre 400 (128MB)
|
|
1024x768 with no AA
|
126.1
|
95.8
|
95.4
|
95
|
94.2
|
85.4
|
|
1024x768 at 4x AA
|
100
|
92.2
|
33.1
|
77.3
|
81.8
|
25.2
|
|
1280x1024 with no AA
|
126.1
|
95
|
95.4
|
89.2
|
89.2
|
58.4
|
|
1280x1024 at 4x AA
|
100
|
61.5
|
33.1
|
44
|
39.2
|
15.5
|
Analysis
Our OpenGL test with this popular title resulted in the scores grouping a bit closer together. The Xabre 400 placed last (again) and it didn't appear to support JK2's use of hardware shaders as the option was grayed out in the game's menu. Without AA, the remainder of our value picks managed to push past 80 fps, making any of them acceptable choices for OpenGL gaming -- especially with "Quake III"-based games.
Increasing the AA level to 4x places serious demands on video memory bandwidth and the two Ti4200-based cards continue to surpass 70 fps at 1024x768. The least expensive card in this comparison, the Radeon 9000 Pro, immediately loses steam when going to 4x AA at either resolution. It wasn't until we moved up to 1280x1024 with 4x AA that we saw a significant drop in performance with the GeForce4 Ti4200-based cards.While there's no denying the power of ATI's Radeon 9700 Pro, it's nearly twice the price of the 128MB Ti4200 from Gainward. If you're using a 17-inch display or smaller, chances are that your in-game resolution is 1024x768 or lower. For these lower-resolution gaming environments, save your money and get a GeForce4 Ti4200 or the Radeon 9000 Pro.
Simply put, most modern video cards are best experienced with a 1.5-GHz processor or greater. Considering that 2-GHz CPUs are quickly approaching $100, you could spend less on a CPU upgrade and enjoy a performance boost with all applications, including your favorite games.
Comments
Add a Comment