Conclusion

Add it all up, and we don't see a whole lot of change for the digital video market status quo, where ASICs and ASSPs rule. Media processors like the Storm-1 should compete well for share in corners of the market where its ease of programming and software reconfiguring are truly compelling. Early stage development of new digital video products where content, standards and usage have yet to settle out. Or better yet, applications where multi-standard is a requirement not for tomorrow but today, for example, video or image transcoding, or where providers offer multiple services like voice, video and data.

Thinking along those lines, FPGAs and more conventional DSPs should have a lot more to fear from chips like the Storm-1 than conventional video ASICs. Like programmable media processors, DSPs and FPGAs promise flexibility through software and reconfigurability (respectively). FPGAs tend to thrive in non-price sensitive applications, and both work well as prototypes, proof-of-concept or get-started vehicles, and in places performance demands are a bit more modest. But in the historically high volume digital consumer electronics market, it's hard to see anything but the absolute most cost-focused solutions in place. Still back to ASICs and ASSPs.

So should such an assessment of Storm-1's market viability come as a shock to SPI? We doubt it. When it comes to naming their key target markets, things like HD DVD players aren't on the company's list. Rather (and at least for now), the company names video conferencing, video surveillance and multi-function printers as key targets. All are good fits according to the aforementioned criteria, and none raise expectations for the kind of volume traditional CE devices like DVD players or digital camcorders do.

Programmable vs. non-programmable: it's different for GPUs. But wait, didn't we just describe how mainstream GPUs not only have made the transition to more elaborate programmable processors, but now are making a move into general-purpose, streaming processors? Then why would a similar approach for video chips mean only minority share, still leaving the lion's share still to less-flexible, cost-focused ASICs or ASSPs? If a programmable solutions works for high-volume graphics, why not high-volume video?

The difference is that for today's GPUs, programmability is a requirement, but for today's mass-market video chips it's a design choice. Thanks to the evolution of gaming, GPU programmability is not just a nice feature or an architectural preference, it's a hard requirement, essential for providing a near-limitless range of application-defined visual effects. A non-progammable GPU today is a non-starter on almost any platform. For a video codec, programmability is a design choice that offers the promise of better flexibility, but it's a design choice nonetheless. There is nothing in the H.264 spec that says anything about the codec having to be programmable.

But inflection points happen, and things can change. Looking back at programmability in GPUs, the inflection point came not when programmable solutions outperformed fixed-function devices on throughput, but when they became fast enough. The transition began when programmable GPUs could paint complex scenes on reasonable resolution displays at real-time frame rates, and do it with a reasonable die size. Once that point was reached, other goals - namely a wider range of more detailed visual effects - became a higher priority than pure speed.

Perhaps in the longer term, the most enticing high volume market for future media processors like SPI's could be displacing a combination of DSP + 3D/2D chip. DSPs today are the rule for digital voice, where soft-programmability is a requirement. And in designing Storm-1, the company did pay attention to optimizing power efficiency.

Carriers are ramping more services demanding multiple standards, as well as features like imaging, 3D/2D graphics and not just video playback but even camcorder support. Take a look at Nokia's N93, for example. And while the jury is most assuredly out with respect to how customers value (i.e. are willing to pay for) such services, the combination of disparate yet concurrent media processing may call for an architecture like SPI's.

Stream Processors, Inc. and its investors are wisely choosing some safer footholds to pursue initially, but clearly must be thinking about a bigger pond for some time down the road. TI is selling a DSP into two of every three mobile phones (src: Databeans), not too shabby. Of course, SPI would hardly be the only fish in that pond, there are already dozens positioned looking for that market to explode, from the big names like TI, AMD, Nvidia as well as a whole host of smaller players.

Like its predecessors in the programmable media processing game, SPI has creative, well-engineered technology. But the market is ruthless, and doesn't necessarily care if something is creative or well-engineered, it has to do the right job at the right price and right power budget. So also like its predecessors, SPI has its work cut out for it, as stretching from its early market footholds to broader markets will be anything but easy.