ASUS Radeon 9800 XT: Turning-point
Benchmarking
| Test configuration: |
|
| CPU |
P4 2,4Mhz (Northwood D1) |
| Mb |
Epox 4PDA2+ (i865PE) |
| Memory |
PC3200 (DDR400) 2x256Mb in the dual-channel node
Timings - 3:8:3:3 |
| Video cards |
ASUS RADEON 9800 XT
GIGABYTE RADEON 9800Pro 256Mb
ATI RADEON 9700Pro 128 Mb
ASUS GeForce FX 5900 Ultra 256MB
ABIT GeForce FX 5900 128Mb |
| OS |
WinXP + SP1 |
| Drivers |
Catalyst 3.8
Detonator 45.23 |
Benchmarking software used:
- 3DMark2003
- 3DMark2001SE
- Codecreatures Benchmark Pro
- Village Mark
- RightMark Video Analyzer v0.4
- Comanche 4 Bench
- Unreal Tournament 2003
- GunMetal Benchmark 1.2S
- X2-The Threat Rolling Demo
- Aquamark 3
Notes: The quality settings in the drivers of the video cards are set to "controlled by applications" by default. In these tests, we defined the maximum quality settings. 10 benchmarking programs (5 synthetic benchmarks and 5 gaming tests) were used altogether. Our standard set of test applications was complemented with new DX9 tests - a benchmark based on the demo version of future game "X2-The Threat", "Aquamark 3" and "GunMetal Benchmark". Along with the ASUS RADEON 9800 XT itself, there was also used a board on the base of GeForce FX 5900Ultra 256MB, also made by ASUS - the fastest of all we got of FX, as well as the GeForce FX 5900 128MB made by ABIT, then RADEON 9800Pro 256Mb by GIGABYTE, and RADEON 9700Pro by ATI.
In the Aquamark 3 benchmark, as many as 15 video cards were used. The test is new, quite specific, and it was really curious to look at the general comparison of performance of various video cards from the viewpoint of this test.
We tried to emulate operation of the RADEON 9800 XT through overclocking a 9800Pro to the clock speeds of 9800XT. On the graphs, this card is referred to as "ATI 9800Pro 256Mb Overclock". The difference in results will allow assessing the gain from using the dynamic overclocking system implemented for the first time in the ATI's produce and the improved memory subsystem.
To make the graphs more amenable for perception, all the ATI lines are shown in bold lines, with NVIDIA's with thin lines.
Synthetic benchmarks:
3DMark 2003
3DMark 2001SE
I am not going to comment on the results for 3DMark2003 and 3DMark2001SE any longer. Simply enjoy in silence.
Codecreatures Benchmark Pro
Remember that the benchmark makes intensive use of shaders of version 1.1 of DirectX 8.1 kit. In the default mode with the antialiasing disabled, we see an approximate parity for all the three cards: Radeon 9800XT and FX5900Ultra, and Radeon9800Pro overclocked to XT clock speeds.
But once we enable the antialiasing, everything changes. A pure victory of NV35 in the person of ASUS FX5900Ultra. As far as I can remember, this test has always been the reign of top-end NVIDIA cards. Now we see that ATI has caught up with the necessary level in the test without AA.
RightMark Video Analyzer
Curiously, it's almost the only test where the results for overclocked 9800Pro fully coincide with those for 9800TX.
Village Mark
A very old and funny test. It can't produce FPS higher than 200 and are regarded as invalid. If we bring in the results for solely today's probationers, then graphics won't work - all straight lines of two points.
The green thin line above the thick yellow stands for the results of FX5900Ultra versus ATI Radeon9800Pro. They virtually coincide. The bright violet line represents the results for the overclocked Radeon9800Pro, and overshadows those for Radeon9800XT. In a month or two, the test results might look like a single point in the 1600x1200 resolution.
Gaming benchmarks
Away with the synthetics. Real gaming tests are much more exciting. We used two games of the DX8 pack - (Comanche 4, UT2003), and three benchmarks on DX9 engines (GunMetall, X2-The Threat, Aquamark 3).
Comanche 4
With the AA disabled, all the lines merge into one :). The benchmarks definitely rests on the lack of the CPU performance. Now, bringing in the results produced with the AA enabled. Even with these, at low resolutions which we simply cut off, all the lines come almost out of the same point. But ATI's leadership is undisputable.
UT2K3 Benchmark Utility
The tests were run only in the High Quality mode - it makes no sense running hi-end cards at poor settings, because even without that the results exceed all limits. You can look through the settings in "Unreal Tournament 2003 as a benchmark". The test utility is available for download from here.
The results are pretty standard, and the emergence of 9800XT hasn't brought any changes to the ranking - NVIDIA takes is up to the mark only at highest resolutions of the Antalus demo.
GunMetall Benchmark
This DX9 benchmark is a separate utility made as an add-on to the game. By default, the image display quality is set to the maximum ever possible. It uses version 2.0 vertex programs and 1.1 version shader programs (isn't it a shortcoming?) We covered this benchmark in detail upon its first use in the review "FX5600Ultra vs Radeon9600Pro". The benchmark includes two demos:
Sure, nothing to comment on. Top-end Radeons rule!
X2-The Threat Rolling Demo
This is a new benchmark made on the new engine of the "X2-The Threat" game by Egosoft. The engine does not use pixel technologies, and that is easily seen to the naked eye. On the other hand, there is actively used the Dot3 Bump Mapping, which leaves indelible impression. In any case, this demo is a DX9 test.
The first test is about the default quality settings. FX5900Ultra is somewhere on par with Radeon 9700Pro.
With shadows and antialiasing enabled, the picture is not that dramatic and even FX5900Ultra leads at 1600x1200. But you are unlikely to call this resolution gameable.
Verdict: Radeon 9800XT takes a lead.
Aquamark 3
You can find out more about this new benchmark in our review "AquaMark3 Benchmark". Briefly: the benchmark was released one and a half month ago, on 12 September on the base of the "Krass Engine" engine that serves as the graphics base for all "AquaNox" series games with gradual complication. The same engine will be in the new RPG "Spellforce" (to be released these days) by Phenomic Game Development (the trial version is available from here).
AquaMark3 is unique in its own way. At least, it's only this version that allows assessing the performance of video cards with all the versions of vertex and pixel shaders that were included in DX8 and implemented in DX9.
Unfortunately, we got only a freeware version of this benchmark that does not allows running a test in all the resolutions - only at 1024x768, but at maximum detail.
So, at "high quality" we again see a complete victory of ATI. The Non Ultra version of FX5900 even lost to 9700Pro,which was somehow unexpected, and the ranking of FX5600Ultra took us by surprise completely! We intentionally included into the the benchmark the results of all the "key cards" released for 2002-2003 and even some older, just for fun. The benchmark is new, quite peculiar and it would be interesting to look at the general performance growth for video cards.
The results are displayed in some conditional score points called Triscore, but if divided by 1000, they just give the average FPS rate for a given video card. That is, 9800XT gave 43 fps on the average, with 4.5 fps for GF2MX.
 |
Top Stories: |
 |
 |
|
MoBo:
| |
|
|
VGA Card:
|
|
|
|
CPU & Memory:
|
|
