This blog post references my cover story, 'Instigating a Platform Tug Of War: Graphics Vendors Hunger for CPU Suppliers' Turf' in the October 13, 2005 edition of EDN.

Intel and Microsoft's unveiling of the AC'97 specification in...err....1977....marked a dramatic shift in hardware-versus-software partitioning for implementing audio on PCs. Prior to AC'97, dedicated audio chips handled both the digital and analog functions associated with audio recording and playback....A/D and D/A conversion, mixing, amplification, sample rate and size conversion, etc. AC'97 shifted the bulk of the binary-based tasks away from the audio chip; initially to hardware circuitry housed in the core logic chipset, over time supplemented and supplanted by software running on CPUs. AC'97 supported six-channel audio at up-to 48 kHz sampling rates, with up-to-20 bit precision. Two-channel audio quality went up to 96 kHz and 20-bit. The follow-on High Definition Audio specification, among other things, boosts audio quality to 192 kHz and 32-bit, across up to eight channels.

AC'97 and HD Audio have put the 'hurt' on sound card vendors, but the add-in audio business hasn't completely evaporated. Companies like Creative Labs have focused their marketing energy on two classes of potential customers; musicians and sound engineers (who value the low latency of a hardware-based audio subsystem approach, along with higher sample rates and sizes than AC'97 allowed; these folks will often also pay extra for a USB- or Firewire-tethered external audio module that moves the analog circuitry outside of the noisy PC environment) and gamers (who value the combination of low latency, rich, high frame rate visuals and immersive surround sound that Creative and its competitors claim isn't possible with a CPU-centric audio implementation). And Creative apparently thinks there's sufficient business opportunity going forward to continue significantly investing in its franchise; less than two months ago the company launched the 3.1 million transistor X-Fi audio processor, along with a family of sound cards (Slashdot discussion and review links here).

Where else could audio processing take place? What about....on the GPU? I was first made aware of this possibility earlier this summer during an interview with Falanx's CEO, Borgar Ljosland, at SIGGRAPH. Further digging uncovered BionicFX (I absolutely love the 'Your video card. More cowbell.' quote on their home page), a series of audio effects algorithms that run on Nvidia graphics cards. Aside from the fundamental challenge of crafting audio DSP routines in such a manner that they 'fool' the GPU into thinking it's processing graphics primitives, it seems that juggling simultaneous audio and graphics functions, both within a single application and between multiple concurrently-running applications, will be tough to robustly implement. But I commend BionicFX for its fresh approach to the audio processing problem.