I have decided to pick two inexpensive graphics cards freely available to put up against the FX5200. First up is the Ti-4200 64MB (this one is an Asus). The budget card of choice for quite some time, I found some Ti-4200 64MB cards on Pricegrabber for as little as $112, with 128MB versions available for the price of a beer more. I also decided to include the Radeon 9000 Pro 128MB, which I managed to find available for just $60, less than half the price of the FX5200 on test today. The newer 9200 Pro is also on offer for around $75, which differs only in (pointless) 8x AGP support and increased clock speeds. There is little point in comparing a budget card like this to the likes of an FX5900 or a 9800 pro because its just not representative. That would be like including a Ferrari Enzo in a station-waggon shoot-out. Test rig is a Pentium 4 2.4 @ 3.3GHz, with 512MB PC3700 running at 275MHz FSB / 220MHz mem.
No doubt some forum backlash will be in order for me for including this old timer in a graphics card test, but I'll think I will let the still-perfect scaling of performance in the various resolutions do the talking. The very fact that Quake III is still one of the most played online games in the world makes this result relevant in its own right.
3DMark 2001SE is a handy test that uses a real game-engine to give a scalable and easily reproducible arbitrary score based on relative Direct X 7 and 8 performance. It's getting long in the tooth now, but is overall a far more useful test than its newer 2003 brother. The FX5200 cannot compete with the Ti-4200 in this test, and is closer to the $60 Radeon 9000 in performance.
Codecreatures is one of the most stressing Direct X 8.1 tests out there and all of the budget cards features had a hard time keeping frame rates smooth. The Ti-4200 is the fastest by a country mile.
Unreal Tournament 2003 is the new king of the hill when it comes to online FPS games, and quite rightly so. Its a great looking and fun title, as well as being a very stressful test on your graphics card. We used [H]ard|OCP's handy benchmarking utility to return these scores. The FX5200 is considerably faster than the Radeon card in this game, but a long way short of the cheaper Ti-4200.
Comanche 4 is one of the least efficient game engines I have ever seen, full of overdraw and frankly awful looking visual effects, this should be a great test for seeing how the FX5200s limited z-compression abilities harm its performance. Sure enough, the FX5200 returns lower than average results, almost neck and neck with the Radeon 9000.
I have brought Power VR Village Mark out of retirement for the day, it's a very limited test by today’s standards but should allow us to see how well the cards deal with overdraw. Villagemark was made specifically to show off the amazing z-compression abilities of the Power VR Kyro based cards (which had tile-based rendering).
3DMark 2003's overall score is a total waste of time, but it does generate a few handy pixel and vertex shader specific tests, as well as some pretty accurate fill-rate information. This test illustrates perfectly where the FX5200 bottleneck is. It has by far the highest single-texture fillrate, but its multi texture performance is considerably slower than the 9000 Pro (a card based on 2-year old technology), as well as its own single-texture performance. This explains why extreme clock speeds are still resulting in a card substantially slower than a lower-clocked Ti-4200.
Take note of the 5200's very strong pixel shader performance relative to the competition, it's a shame it doesn’t not have the architecture to put them to good use.
Performance with IQ on
IQ settings (Full Scene Anti Aliasing and Anisotropic Filtering), in general, are a luxury top-flight graphics card like the 9800 Pro and 5900 Ultra buy you. However, they can still be enabled on all of the entry-level graphics cards on test today).
As you can see, in Quake III it’s (just) about possible to play with 1024 x 768 with all the candy up plus IQ on all of the cards here. 1280 x 1024 means that it becomes a bit jerky at times, and 1600x1200 is unplayable across the board.
In Codecreatures, a test where complex texture handling and shader utilisation is high, you can once again see the FX5200 drop to Radeon 9000 speeds. The Radeon's drivers won’t allow 4 x FSAA / 4 x Aniso at resolutions above 1280 x 1024 and automatically drops FSAA down to 2x. For that reason I have removed these results. The fact that the aging Ti-4200 is nearly as fast at 1600x1200 as the FX5200 is at 1024x768 will make for uncomfortable reading for nvidia.
