Previous Page > Specifications, Overclocking and Test Platform > Page 1, 2

3DMark2001SE

Futuremark's 3DMark2001SE still has a valid role in the testing world as it is based on the DirectX 8.1 technologies which are still used heavily it today's gaming world. It also doesn't have the stigma of the new 3DMark03 because it was based around the Max Payne game engine. The scores included from the benchmark are the overall, single and multitexture fill rates:

The overall scores for the ATI Radeon 9600 Pro are nothing really spectacular. Many of the previous generation nVidia GeForce4 cards are able to get higher scores in these tests. At the stock speed of 400/300, the multi-texture fill rate is very close to the specification fill rate. What is impressive is the jump in the scores from the overclocking of the processor. The scores jumped up 17.9%, 20.7% and 32% respectively.

3DMark03

There has been a lot of controversy regarding FutureMark's 3DMark03, but unfortunately there are no available DirectX 9 games available yet for testing. As a result, it is the only real benchmark available to test the latest shaders built into the DirectX 9 video cards. I've included the overall result for completeness, but the numbers that really matter are from the individual technology tests.

Much of the results with 3DMark03 reflect the same as the 2001SE results. It was interesting to see that the fill rates for both single and multi-texture were lower than those reported by 2001SE, but once again the multi-texture results was not that far off from the specification. The most interesting result is the Pixel Shader results that showed a very smooth looking 28.2 frames per second. Overclocking results were once again impressive as with 2001SE showing between a 20 to 31.7% increase over the default speeds.

Unreal Tournament 2003 Benchmarks

Unreal Tournament 2003 is one of the biggest DirectX 8.1 games currently available. For this benchmark, I used the UT2k3 Benchmark Utility version 2.1 from HardOCP to run the tests. All settings in the benchmark were set to high detail and run on the DM-Antalus flyby. The test was run at 5 different resolutions with no anti-aliasing or anisotropic filtering turned on as well as the 4x quality setting for both and at the stock and overclocked speeds. Results reported are the average frame rates from the benchmark report:

Looking at the results between the various resolutions, performance of the 9600 Pro was very predictable with an approximate drop of about 30% for each increase in resolution. The only major dropoff occurred at the 1600x1200 setting with AA and AF turned on. At this point, the decrease was about 50% of the 1280x1024 resolution. Using the default clock rate, smooth playable frame rates were achieved up to 1280x1024 without AA and AF and 1024x768 when they were enabled. Overclocking improved overall performance allowing for smooth frame rates to be achieved at 1600x1200 without AA and AF and just barely 1280x1024 with them on. Performance at higher resolutions was around 25% higher using the overclocked settings.

Return to Castle Wolfenstein

Return to Castle Wolfenstein is a fairly old game now based upon the Quake 3 engine. A good number of games still use modified versions of this engine making it still valid for benchmarking. Benchmarking was achieved using the Q3Bench application with max detail and 32-bit color and textures set. All tests used the checkpoint.dm_60 demo map with both no anti-aliasing or anisotropic filtering and with both set to 4x quality mode:

Since Wolfenstein uses the older Quake 3 engine, the game is much more limited by the CPU than the video cards. As a result, we don't see any appreciable drop off in performance due to increasing resolution until we either turn on AA and AF or use the highest resolutions possible. Once either of these conditions occurred, performance dropped about 30% with each increase in resolution. At the default clock speeds, the game is very playable at the highest resolution and is even playable with AA and AF set at the 1280x1024 resolutions. The amazing part is that when overclocked, the game is still fully playable with full detail settings including AA and AF even at the 1600x1200 resolution!

Conclusions

Overall the Sapphire Atlantis Radeon 9600 Pro is a very good performer. When taking into account the cost of the card and the overclocking potential, it becomes and even better value. The one big caveat with this board is the anti-aliasing and anisotropic performance. While it is possible to use these functions to improve image quality in 3D games, it is generally only possible at the lower resolution settings of 1024x768 or lower. This is due to the roughly 40% decrease in performance by turning these features on. The board is not capable of supporting the functions at higher resolution because of its limited memory bandwidth. If high resolution, high quality gaming is that important, then one should look at the more expensive high-performance boards. Of course, most gamers are not striving for this, and that is what makes this card perfect for less discerning gamers.

The big test for all of the DirectX video cards still awaits. Only when the major DirectX 9 games like Doom III and Half-Life 2 are released will the true performance of the cards become known. Those looking to play these games may be well served by waiting until good solid benchmarking results can be generated with these games. Only then we will really know how well these boards perform. Of course, the other advantage is that by the time the games do come out, the prices of the video cards will also likely have dropped.

Manufacturer's Site