To the desktop musician, life is an endless quest for more horsepower. No matter how fast our computers are, we always want them to chew through audio processes more quickly, provide a higher track count in our DAWs, let us run more instances of our favorite reverb plug-in, and enable our software synthesizers to give us more polyphony. Fortunately, computer processing power is increasing at a remarkable pace.
New generations of both Macintosh and Windows CPUs have hit the market in recent months: Apple released the new G4 series, and PC users can now choose between Intel's Pentium III processor and AMD's impressive new Athlon. Naturally they're faster than their predecessors, but each also offers new features that can greatly enhance music applications.
MUMBO JUMBOTo see what all the fuss is about, I surfed over to the chip makers' Web sites and devoured every press release and white paper I could find. I learned that the new chips sport such practical features as a 9-issue superpipelined, superscalar 80x86 processor microarchitecture and source-synchronous clocking (clock-forwarding) technology. When I read the news about point-to-point topology and high-speed backside L2-cache interfaces, it was easy to see why musicians were getting excited. Okay, I'm fibbing-it wasn't easy at all. To be brutally honest, the only thing I really got out of all that gobbledygook was a naughty chuckle at the word backside.
With all due respect to the supremely nerdy, only a few important specifications separate one generation of processor from another. Clock speed is the most familiar. Each new generation of chips supports higher clock speeds than its predecessor; the industry is now starting to measure clock speed in gigahertz rather than megahertz.
But higher clock speed results in only a modest increase in processing power. To really speed up a computer, you need to change the way that a chip handles data. For example, when Intel replaced its older 16-bit processors with the 32-bit Pentium, the company altered the pipeline so that the chip could perform more operations in parallel (at the same time). Even though the Pentium had clock speeds similar to those of its precursors, it could perform more simultaneous operations and consequently complete the same number of operations in less time.
Cache design is another area in which new chips soar past their forebears. When they were first introduced, Intel's affordable Celeron chips were dogs because they had no onboard cache. Consumers balked at the Celeron's poor performance, so Intel quickly added a small on-die cache that ran faster than the cache on a Pentium II chip. The Celeron's resulting improvement almost rendered the more expensive Pentium II superfluous.
Expanding a processor's instruction set also significantly affects its performance. For instance, Intel's MMX extensions gave multimedia developers new and more-efficient ways to accomplish certain processor-intensive tasks. Applications written to take advantage of the new instructions offered superior performance. Music-application developers were intrigued by MMX's parallel processing, but they were put off by the fact that both the Pentium and Pentium II processors could not simultaneously use MMX integer math and the floating-point math employed by most audio programs. As a result, MMX is rarely used in music applications. The Pentium III processor avoids this issue altogether by dealing only with floating-point calculations.
NITTY GRITTYThe Pentium III, Athlon, and G4 feature all sorts of improvements that make them faster than their predecessors, and we power-hungry desktop musicians reap the benefits in everything we do. Of all the advancements, however, the new processor instructions are what's bringing the brightest smiles to the faces of developers and musicians. Programmers on both platforms now have access to processor-level functions that speed up those all-important floating-point calculations.
The new instruction set is called Single Instruction Multiple Data (SIMD). For the Pentium III (see Fig. 1), Intel has named the feature Streaming SIMD Extensions (SSE). In simple terms, SIMD allows programmers to issue one instruction that will be performed on several different items-it's akin to saying "Feed the pets" instead of "Feed the cat, feed the dog, feed the fish," and so on. SSE applies the MMX's parallel processing to floating-point functions.
Software synthesis is one music application that benefits greatly from SSE. With a Pentium III, Yamaha's S-YXG100 software synthesizer supports twice as many XG wavetable voices as it does with a Pentium II. Its VL virtual acoustic-modeling synthesis is monophonic with earlier chips; with a Pentium III it reaches 8-voice polyphony, due to the newer processor's SSE features and increased overall speed. Other manufacturers also have implemented the Pentium III's SIMD extensions to increase their software synthesizers' polyphony.
As you might infer from its name, SSE facilitates the streaming of multimedia over the Internet; music applications developed for that purpose get a performance boost with a Pentium III. Seer Systems (maker of the Reality software synthesizer) offers SeerMusic Player, a free browser plug-in that uses SSE to make high-quality multichannel audio available more quickly and easily. The latest version of Beatnik's Beatnik Player, a browser plug-in that supports sampling and has a built-in software synthesizer, is also optimized for SSE.
Digital audio mixing and DSP plug-ins eat up floating-point horsepower in a hurry. They, too, stand to benefit from the new processing instructions. For example, Digidesign's recently released digital audio workstation, the Digi 001, relies on the computer's CPU for mixing and effects processing. Whereas Pro Tools has always depended on TDM plug-ins running on custom DSP hardware, the Digi 001 features a new host-based plug-in format called Real Time AudioSuite (RTAS). This means that tons of floating-point operations get dumped on your CPU. Digidesign has SSE-optimized the Windows 98 version of Pro Tools LE to ensure that you get the most from your host.
BY ANY OTHER NAMEAdvanced Micro Devices has always marketed its K6 processors as a low-cost alternative to Intel chips. However, many desktop musicians have shied away from AMD chips due to the purported inferiority of the K6's floating-point unit (FPU). For the same reason, some manufacturers of plug-ins and software synthesizers still caution their customers to use only true Intel chips.
The latest K6 processor, called the K6-III during development but now known as the Athlon (see Fig. 2), decisively resolves this FPU issue. AMD redesigned its FPU for the Athlon, and the results are impressive. The improvements make floating-point calculations cleaner and applications more stable.
The Athlon uses its own version of SIMD enhancements that AMD lumps under its 3DNow banner. (3DNow is a broad category of special chip functions designed to enhance 3-D graphics performance and other multimedia streaming applications.) Because AMD's implementation of SIMD is different from Intel's, software applications have to be adapted specifically to each chip manufacturer's instruction set, increasing the time and cost of the software development process. But now that the Athlon has emerged as a worthy competitor, developers are more willing to cover the extra expenditure needed for adaptation.
Another nontechnical benefit that AMD provides for musicians is the price pressure it brings to the PC chip market. Although past K6 processors took a backseat to Pentium chips in some respects, their low cost forced Intel to think in terms of value. Consequently, Intel developed the less-expensive Celeron processors, the latest of which have become quite popular among desktop musicians. And because the Athlon can compete with the Pentium III in terms of performance, the bang-for-the-buck factor tips the scales still further in the desktop musician's favor.
BELIEVE THE HYPEApple boasts that its new G4 Macs (see Fig. 3) are the first-ever personal computers to offer "supercomputer-level performance," based on the new processor's ability to carry out more than one billion floating-point operations per second (gigaflops). The G4 incorporates a new feature called the Velocity Engine. Whereas previous chips dealt with floating-point operations one at a time, the Velocity Engine enables the G4 to handle up to four 32-bit floating-point operations simultaneously.
Maybe it's me, but "Velocity Engine" is one of those Madison Avenue names that immediately make me suspicious. When I first started reading breathless press releases about the new "G4 with Velocity Engine," I thought it had to be the brainchild of the same marketing team that long ago gave us "GI Joe with Kung-Fu Grip." And when I saw that the G4's processing capability is measured in gigaflops, I became extremely wary of the hype.
Fortunately, the Velocity Engine lives up to its publicity. It significantly boosts the performance of floating-point-intensive operations such as synthesis and DSP. For example, BitHeadz has optimized its Unity DS-1 software sampler and Retro AS-1 software synthesizer for the Velocity Engine. As a result, both products offer twice their previous polyphony. The improvement also allows more-sophisticated playback algorithms and higher-resolution effects.
As with the Pentium III's Streaming SIMD Extensions and the Athlon's 3DNow enhancements, the G4's Velocity Engine is a boon to digital audio mixing and effects. Version 5.0 of Digidesign's Pro Tools and Pro Tools LE have been optimized to take advantage of the new Mac chip's efficiency.
You may not know or care about floating-point calculations, but as desktop musicians we use them constantly. They are a vital part of digital mixing, DSP effects, and software synthesis, all of which are near and dear to our hearts. Luckily, the major chip manufacturers for both platforms are making great strides in speeding up our favorite processes. Any way you slice it, that's music to our ears.
Brian Smithers is a musician, conductor, and arranger at Walt Disney World. Share your thoughts with him through his Web site at http://members.aol.com/notebooks1.