|
Our second set
of tests were run with Futuremark's controversial "DirectX
9" benchmark, 3DMark03. 3DMark03 is similar to its
older cousin, 3DMark2001, but it uses a mixture of DX9 and
DX8 pixel and vertex shaders to assess a video card's
performance. This test is meant to be run only on "DX9
Class" hardware only, which is why the GeForce 4 Ti4600 is not
represented here.
 |
|
Benchmarks
With 3DMark03 |
|
The
Controversial Synthetic DX9 Benchmark |
|
9600 PRO
9500 PRO
9700 PRO
In the 3DMark03
tests, the Radeon 9500 Pro outperformed the 9600 Pro by
approximately 10.5%. Obviously, the Radeon 9700 Pro
had an even larger performance advantage, where it outpaced
the 9600 Pro by over 47%. Something we found peculiar
was the CPU Score. The Radeon 9500 and 9700 Pro performed at
virtually identical levels, but the 9600 Pro was 400+ points
behind. We suspect driver optimizations may change
this unusually low result.
|
|
|
Head-to-Head / Performance
With
Comanche 4 |
|
Looks
Like Recent New Footage |
|
We continued our
DirectX testing with some benchmarks from Novalogic's Combat
Helicopter simulator, Comanche 4. Like 3DMark2001, Comanche
4 makes use of DirectX8 Pixel and Vertex shaders, and is a
useful tool for testing performance under DirectX 8. This
benchmark generally scales more with increased CPU speed, so
don't be fooled by what looks like low frame rates.
From this point
forward, you'll notice there are quite a few "N/As" listed
in our graphs. These are there because the GeForce 4
Ti4600 is not capable of 6X AA and the Radeons could not run
some of the tests with the specified settings. At
1024x768 with Antialiasing disabled, the Radeon 9600 Pro
didn't do so well, actually it was the lowest performing
card in the group. When AA was enabled however, the
9600 Pro outpaced the GeForce 4 Ti4600 using all AA methods
at both resolutions. We saw an extremely sharp
drop-off in performance, when we enabled AA at 1600x1200
with the 9600. The scores were much lower than we
expected with this card, and suspect the results may change,
again with a future driver revision.
Some Quake 3 & UT2003 Benchmarks
| 
