CPU utilization
We'll kick things off with a set of CPU utilization tests using RightMark 3D Sound. The ADI 1988A only supports up to 32 voices in this test, so we don't have results from it at 64 or 128 buffers.

On the Athlon 64 X2 3800+, there's a sizable gap in CPU utilization between the Oxygen-based cards and their competition. That gap grows dramatically as we increase the number of concurrent buffers, while the X-Fi's hardware acceleration allows it to maintain a reasonable level of CPU utilization, even with 128 buffers.

Note that there's little performance difference between the Auzentech and Sondigo cards; they use the same audio chip and drivers, so they're likely to offer all-but-identical performance in 3D audio tests. It's also interesting to see that the ADI 1988A has much lower CPU utilization than either Oxygen-based card, at least up to 32 buffers. Analog Devices uses the Sensaura 3D positional audio library, and Sensaura is, incidentally, now owned by Creative.

Overall CPU utilization drops a little when we move to the faster Athlon 64 X2 5000+, but the Oxygen-based cards are still occupying far more CPU cycles than either the X-Fi or our ADI onboard audio.

CPU utilization with digital output encoding
Since real-time Dolby Digital Live and DTS encoding are key facets of the Oxygen HD's appeal, we decided to run a quick CPU utilization test to see whether the encoding process saps CPU cycles. The results below come from the X-Meridian, but the Inferno's performance was essentially identical. The card was tested with standard analog output and then with digital output pushing DTS Interactive and Dolby Digital Live bitstreams.

There does appear to be a computational cost associated with on-the-fly encoding, but it only becomes significant when we move beyond 32 buffers. RightMark 3D Sound's EAX test doesn't show much difference between the output methods, either.

A note on 3D positional audio quality
We've encountered dodgy EAX implementations in the past, notably with certain versions of Realtek's HD audio codec drivers, so I fired up RightMark's positional audio test to confirm that the Inferno and X-Meridian don't cut any corners with 3D audio. Both cards correctly place 3D sounds, and they properly honor EAX occlusions and obstructions. So do the ADI 1988A codec and X-Fi Fatal1ty, and without dipping into the X-Fi's support for more advanced EAX features, my ears can't detect any real difference in the positional audio quality across all four configurations. However, once you start tapping the X-Fi's support for a greater number of 3D sounds, you start to hear things that you don't on the other cards—most dramatically in Battlefield 2.

Unfortunately, the quality of in-game audio isn't as high as we'd like for playback quality testing. Most games use compressed audio, and that makes discerning the already subtle differences between each card's playback quality difficult at best. The dynamic nature of games also makes conducting reliable blind listening tests extremely complicated. However, after spending a lot of quality time benchmarking the cards, I got a good feel for how each sounds in games. The cards really do sound similar, but gunfire on the X-Meridian is noticeably louder than it is on the other cards. Also, when running with fancier EAX effects, the X-Fi seems to provide a slightly richer, more immersive sound. I'm not entirely sure whether that's due to superior playback or the addition of other background sounds, though. Running the X-Fi with the same in-game audio settings as the others doesn't convey quite the same effect.