generated quite a bit of buzz immediately after its release.
Certain parties deemed it was not a
meaningful benchmark because there aren't any games that use
the same shader programming techniques. Whether it's
indicative of actual game performance or not, the bottom line is that it remains one
of the only tools we have to test DirectX 9 class
pixel and vertex shaders in a simulated game environment.
Until true DX9 class games arrive that give us the ability
to accurately test performance, 3DMark03 is "it". We
ran 3DMark03 on all of the cards represented here, at the
benchmark's default settings (1024x768x32).
It's What We've Got For Now
RADEON 9800 PRO 256MB
RADEON 9800 PRO
RADEON 9700 PRO
GEFORCE FX 5800 ULTRA
As expected, the 256MB Radeon
9800 Pro outperformed all of the other cards in every "Game
Test". The overall 3DMark03 score showed a paltry 3.8%
performance gain over a 128MB Radeon 9800 Pro, but breaking
down the individual "Game" scores tells the whole story.
In the least demanding "Game Test 1 (GT1)", the 256MB Radeon
outran the 128MB model by less than 1%. In "Game Test
2 (GT2)" and "Game Test 3 (GT3)" however, the results were
better, with the 256MB card pulling ahead by about 4.8% and
3.0% respectively. In the most demanding test in the
suite, "Game Test 4 (GT4)", the 256MB Radeon 9800 Pro showed
the largest performance gain over the 128MB card, 8.6%.
When compared to the 9700 Pro and GeForce FX 5800 Ultra, the
256MB 9800 Pro's advantages are even larger.
We continued our DirectX
benchmarking with Novalogic's combat helicopter simulator,
Comanche 4. Comanche 4 uses DX8 class pixel and vertex
shaders to produce some of the realistic visuals used
throughout the game. Don't be distracted by what look
like relatively "low" frame rates in this test, this
benchmark is heavily influenced by CPU performance and
overall system bandwidth...
At 1024x768, regardless of
whether or not Antialiasing or Anisotropic Filtering were
enabled, the 256MB Radeon 9800 Pro performed at virtually
the same level as the other cards. It was slightly
faster in most configurations, but not by much. The differences
in performance at 1600x1200 were much more dramatic,
however. Without AA at 1600x1200, all of the scores are
basically identical, but with 4X AA enabled the 256MB Radeon
was approximately 4.7% faster than the 128MB model.
With 6X AA enabled the 256MB 9800 Pro surged ahead with a
14.5% advantage, and when 4X AA and 8X Anisotropic Filtering
were used together it jumped out to an even larger 16.7%
UT2003 & Quake 3 Benchmarks