The A3D 2.0 engine and API has been undergoing progressive tweaking by Aureal.  With the release of the latest drivers we decided to take a quick look  at where performance was with the initial release of A3D 2.0 and where it is today using Heretic II as our prime the benchmarking tool.  Hertic II is the focus because it uses the latest A3D 2.0 engine that we have access to.

The important file in the driver set with respect to A3D 2.0 performance is A3DAPI.dll and we used the following versions to test the progress in performance.

So lets cut to the chase.  Heretic II has a built in timedemo feature just like Quake II, however it does not come with a prerecorded demo.  I am using a demo called "NTDEMO" provided to me by David Gasior during his tenure with Aureal.  There is nothing special about this demo other than its goes out of the way to make use of the A3D 2.0 features so its an excellent test demo for our purpose.  To use it, you need to create a demo folder in your base directory and copy the demo file into it.    From the console, you set timedemo to 1 and then run "Demomap ntdemo"  All tests were repeated a minimum of three times and the system was rebooted between driver changes just to be safe.  You can download a 120 KB zipped copy of the demo.

All Heretic II and Motorhead tests were run on a BX motherboard with a Celeron 333a oc to 416 Mhz with 64 MB PC100 Ram and  Windows 98 with a 12 Meg Creative Labs Voodoo 2 board using the latest reference drivers from 3dfx at 640*480 3dfx opengl, non-paleted textures.  The latest EP patched version of Heretic II was used and audio was set to high quality for all tests.  All tests are also run with the 2031 beta reference Vortex 2 drivers from Aureal except for the v.2.22 A3D API.dll which was run with 2036 reference drivers and  v.2.25 A3DAPI.dll  which was run with 2040 reference drivers.

The results:

As you can see, it very clear that Aureal is making positive strides to reduce CPU overhead of the wavetracing engine.  In Heretic II, the performance hit of using the full feature set of A3D 2.0  vs. Direct Sound has dropped from 25.4 percent all the way down to just over 1 percent on a Celeron 333a oc to 416 Mhz!  Using Aureal's new optimized Vortex 2 chip on the SQ2500 board we see even better numbers with basically no imact resulting from wavetracting. More more on this new board and compartive scores you can check out our Aureal Vortex 2 SQ2500 Review.

We also looked at the two original A3D 2.0 titles, Motorhead and Half-Life.  Both of these titles use older versions of the wavetracing engine which apparently are not affected by the driver improvement from Aureal.  In the case of Motorhead, the built in benchmark feature is also a possible factor as it really does not make the same use of the wavetracing engine as the actual game.  Having said that, the numbers are constant through all versions with A3D 2.0 performance at  41.8 fps (based on 3 runs of the test) with A3D 1.x performance at 47.6 and no sound at 54.5 fps.  We can take this as an example of an older version of the A3D 2.0 engine running at about a 12.2 percent performance hit.  (Motorhead was version 3.0)

Half-Life is another interesting example which has posted challenges for both EAX and A3D 2.0 implementation.  For most titles the default sound engine is replaced by the A3D 2.0 engine to get optimal support.    Half Life is a special case where they had features the A3D 2.0 engine could not replicate (e.g. software reverb) and the A3D 2.0 engine was instead integrated into Valve's software sound engine.  In addition, it also uses an older version of the A3D 2.0 SDK which is much less optimized for geometry processing than the current engine.  The result is once again, no change in performance with the different A3DAPI.dll version.   This translates to a 25.8 performance hit compared to default audio with no 3d sound or perhaps a more appropriate comparison would show a 13.4 percent performance hit compared to A3D 2.0 with geometry processing disabled but 3d sound enabled.  You can get more details and comparisons with lots of other cards from our Half-Life Benchmark Page.  You might also be interested in reading our ZD Audio Benchmarking page. Just keep in mind that it does not look at wavetracing or the A3D 2.0 engine.  However, it looks at Direct Sound and Direct Sound 3D performance with a number of cards.

So what's the conclusion?  If you can accept that Heretic II with its more recent A3D 2.0 engine as the right program to make this comparison then Aureal is clearly headed in the right direction in terms of reducing the CPU usage of geometry processing for wavetracing down to insignificant levels.   The trick is to keep it that way in future titles and there are some signs that its not as simple as using the latest wavetracing engine.  For example, our Unreal Tournament performance benchmarks show that the demo is showing a more significant performance hit for wavetracing than Heretic II. The same is true for Descent 3.  What we need and want to see from the Vortex 2 wavetracing performance is imacts similar to what we see in Heretic II, and not what we have in Half-Life and the Unreal Tournament demo.   Aureal is definately moving in the right driection but we want to see it at the Heretic II level or less on all future titles.