Quality aside, what's the real effect on hardware-based sound on CPU use, aside from the FPS performance in games? Can it be measured, and what are the results? We know that in games, higher frame rates can be obtained in some cases, but how much do you gain when playing audio? Here we go with it...
We ran Creative's X-Fi Elite Pro as well as ExtremeGamer Fata1ity (both have the optional X-RAM included) on two platforms: one is a three-year old 3.2GHz single-core Pentium 4 "Northwood" FSB800 on Intel D875PBZ mainboard and Nvidia Quadro 980XGL graphics card; the other the 3.33 GHz quad-core Intel Xeon 3220 FSB 1667 reference system using Asus' Striker Extreme mainboard. These we compared against the integrated sound - OK, not really integrated, Asus provides its Supreme FX ADI988-based audio on a riser card to reduce the noise interference from the mainboard. In fact, this entry is among better ones in the integrated sound field. So it was a good candidate to compare against.
The old Intel board had an earlier version of the Analog Devices Audio, the SoundMax 4XL using ADI1985 codec.
The ouput devices included Creative's brand new Aurvana DJ headphones, an achievement in itself - how did they make such large hi-fi headphones feel so light? - and the Gigaworks 5.1 speaker set, where the biggest challenge was the cable mess.
The round of benchmarking this time included playing various bit-rate MP3 files (up to 450 kbit/s) with effects like virtual surround from stereo to 5.1 and sped-up playback, as well as DVD-Audio playback at 24-bit 96 kHz rate, usually far more demanding than plain CD audio playback. I measured the CPU utilisation in Task manager during the play, both the peak rate and the spread across the four cores. There were no other active apps or applets.
With X-Fi, I used Creative's Mediasource Player and its utilities for sound enhancement, checking CPU usage in Task Manager. Mediasource DVD-Audio player did DVD-Audio playback using Creative's own DVD-Audio sampler from the old Audigy 2 (they don't seem to give DVD-Audio sample discs these days).
When playing the usual 128 kbps MP3 files, no matter what effects are applied, the CPU utilisation on the quad-core Intel never passed one or two per cent of just one of the four cores - all others were idle. Even when going to 384 kbps files, the CPU utilisation remained exactly the same. In DVD Audio, that jumped up to about 3 to 4 per cent - again, in one core out of four.
On the older machine, when equipped with X-Fi, the CPU utilisation was around two to three per cent on that single core, and between five and seven per cent when playing DVD Audio.
When using the integrated audio of quad-core Striker Extreme, and playing via Micro$oft Media Player 10 in plain stereo mode, without effects (the integrated audio applet didn't allow for any specific effects), the basic CPU utilisation in standard stereo playback was between one to two per cent, but across all four cores, and leaning more towards two to three per cent when going for 384 kbps files. Kudos to MS for making Media Player multi-threaded, but all the same, why spread a simple 320 kbps MP3 stream playback across four cores? What would that mean in gameplay, when all busy cores are constantly interrupted by the sound-playing thread?
For the old system's integrated audio, the results were even more significant - the 320 kbps MP3 playback through Media Player reached 12 per cent CPU usage at times, and 384 kbps file topped 14 per cent! This is fine if you're just listening to the music and doing nothing else (OK, maybe a virus scan), but serious CPU-intensive apps performance would be impacted.
In summary, if just playing music, a card like Creative X-Fi - any of the models - will save you some CPU time, but more importantly, provide up to 116 dB s/n ratio and a vast array of hardware-accelerated effects to enhance your song, like sound cleanup, proper stereo to 5.1 or such expansion, crystaliser and so on, all that without any CPU penalty. Your, say, 3-D building rendering or Photoshop conversion will not be slowed down a bit, if you like hi-fi grade musical impressions during your design work, the quality is there. It will be very interesting to compare both the performance and sound quality of X-Fi with the new Asus Xonar when it appears in the coming month, look towards a detailed comparative review right here.
The performance benefits are far higher in the games. I've done early benchmarking in Quake IV using the Sparkle Caliber P880+. Even on this highest-end 3.33GHz quad-core with top-notch GPU, I'm talking about a five to six per cent FPS net gain - it would vary in different games with the amount of sound effects that you use, like for instance the Creative EAX location-bound spatial effects, where the echos and reverberations are adjusted to the location inside the game. You can't even think of doing it with "software" unaccelerated mainboard sound.
Anyway, that net speed gain is equal to pushing the CPU by another whole speed grade - like 3.33 to 3.66GHz for instance, but without the heat and stability penalties of that higher speed grade - if you can get to it, at all.
That's why, all these years, I was Ballmer-style "whispering" into mainboard vendor's ears that, for a high-end entry, you either put proper hardware-accelerated sound on board, or you may as well not put any sound at all, and use the space for something more useful. Like, say, another two fan connectors? Or, better still, another 4-pin standard power input connector from the PSU - either EPS or old storage drive type - to accomodate the extra 12V strength for that extra GPU or CPU. After all, the boards are getting power-weak trying to support quad-core CPUs and dual GPUs with extra hot chipsets and RAM to boot, all at the same time.
The irritating little useless on-board software sound thingie stands in the way, in these times when high-end mainboard real estate is becoming as precious as the "real" high end real estate.