Originally Posted by triggerc
What part of "Audio Processing Unit" is so hard to understand? If you want to argue semantics, fine, the Envy24HT-S is host based, it doesn't count as a bonafide processor, but the CMI8788 is built from the audio processor up. I fail to see how the EMU CA20K is more of a "processor" than the CMI8788.

BTW I'm still laughing, especially at the following bolded section

Wow, I think I must relearn everything I've ever known about sound cards. They don't process the audio? I don't think I'll need a sound card anymore since it doesn't really do anything, I'm gonna have to buy X-Fis again because they ACTUALLY PROCESS THE AUDIO, unlike other sound cards. /Sarcasm

Originally Posted by
The X-Fi architecture
Before exploring where the XtremeMusic fits into the Sound Blaster Xtreme Fidelity family, let's take a moment to explore the 51-million transistor X-Fi audio chip that's at the heart of Creative's new Sound Blasters. Manufactured using 0.13-micron process technology, the chip has roughly half the number of transistors of an Athlon 64 and more than 11 times that of the Audigy, so it's quite a leap from previous generations.

The X-Fi's processing power is divided between five internal units: the sample rate converter, digital signal processor, and mixer, filter, and tank engines. Much of the X-Fi's muscle ripples through a sample rate converter (SRC) that Creative claims pushes over 7000 MIPS. The SRC is actually made up of 256 individual sample rate converters, all of which tackle sampling rate conversions in the same manner. First, the sampling rate of an incoming audio stream is doubled. Next, a poly phase Finite Impulse Response (FIR) filter is used to produce a sampling rate four times greater than the desired output sampling rate. Finally, the sampling rate is reduced by a factor of four for output. According to Creative, this process is nearly transparent, and any loss in quality during sample rate conversions is miniscule compared to the noise generated by even the best DACs available on the market. If you're not convinced, the SRC can be bypassed when it's not needed.

Although the X-Fi's sample rate converter has significantly more processing power than the rest of the chip, it's still only one of five main chip components. The next X-Fi component of interest is the Quartet DSP. Quartet, in this case, refers to the fact that the X-Fi's digital signal processor is made up of four SIMD (Single Instruction, Multiple Data) hardware threads. Each of those hardware threads has two data paths, leading Creative to describe it as TIMD, or Thread Interleaved Multiple Data. Giving each DSP thread dual data paths is a clever way to deal with audio data, which generally arrives in multiples of two. With four "stereo" hardware threads, the DSP can tackle eight data streams at once—perfect for an eight-channel sound card. As one might expect, the Quartet DSP's instruction set is audio-centric. It can handle both fixed and floating-point data types, and Creative claims that the interface is programmer friendly.

The Xi-Fi's SRC and Quartet DSP are undoubtedly the stars of the show, but the chip has several other essential components worth mentioning. More than 1200 MIPS of processing power are dedicated to the X-Fi's mixer, which handles the scaling, combining, and, of course, mixing of audio streams. The "Tank" engine handles all of the X-Fi's delay-based effects, including reverb, chorus, reflections, and inter-aural time delays, while a filtering engine dedicates a couple hundred MIPS to environment modeling, equalizers, and positional 3D audio. The X-Fi also has a transport engine that interfaces with onboard memory and an I/O bus, such as PCI. As you might expect, the chip also has an audio I/O component.