Multimedia Processing

When NVIDIA acquired PortalPlayer, the two had very different architectures for multimedia processing. The former used extended Tensilica cores while the latter used a DSP attached to fixed-function units. The former is just like AMD, while the latter is much more similar to Texas Instruments’ approach. And it turns out that the DSP approach won. Of course, we presume quite a few implementation details changed.

On the other hand, NVIDIA’s in-house Image Signal Processor is what’s being used in the APX 2500, as it was more advanced than PortalPlayer’s. It supports a superset of the functionality of the GoForce 5500’s ISP, which seems to be very competitive with Texas Instruments’ on paper. In practice, these things are all about implementation details so they’re hard to compare accurately without non-public data.

As mentioned above, NVIDIA made aggressive use of many low-power techniques, including power islands/shutoff. They’re obviously not the only company to be using power islands, but not everyone does so either because they’re hard to do. The architecture of the APX 2500 was designed so that every kind of usage scenario (listening to music, taking a snapshot, encoding a video, web browsing, etc.) would only require a minimal amount of transistors to be woken up.

As you’d expect in this kind of design with power islands, everything but a tiny bit of logic waiting for input is powered down when the handheld is on standby. This should assure that the standby power is class-leading despite the use of higher-leakage gates in parts of the chip.

For those interested in a video demonstration of the interface, there's one over at YouTube.

Conclusion & Opinion

Overall, the NVIDIA APX 2500 feels like a very impressive design. The multimedia functionality is certainly class-leading and the power consumption is extremely low. The only downside is that the ARM11 still can’t match a Cortex-A8 despite the high clock speed; but this is already nearly twice as fast as the iPhone’s CPU, and we honestly don’t think it really matters for any realistic usage scenario.

We don’t have any real technical concern with this architecture, although as always in the handheld world we’d love to have even more raw data to draw our own conclusions. But because we don’t have that, it certainly wouldn’t be fair to declare anyone to have a ‘winning’ architecture at this point, despite all the evidence pointing in that direction. We’ll let the design wins speak for themselves in the months and years to come.

This announcement also brings up another question: where does the industry go from here? After all, do we really need more than 720p video, for example? We probably do, as the difference between 720p H.264 Baseline and 1080p H.264 Main/High Profile is very substantial if streamed on an external HDTV, and 4G networks might be able to handle very high bitrates indeed. However, it is doubtful that such higher-end capabilities will ever become so mainstream as to be expected in all smartphones.

There are a few other (possibly more disruptive) capabilities that could also be added to application processors and multimedia companion chips however. AMD has proposed voice recognition, for example, and we’d tend to believe they’re not the only ones looking into accelerating that functionality.

Beyond battery life, there also are two other areas where continuous improvements might remain very obvious to the end-consumer and reviewers in certain parts of the market, and these are the Image Signal Processor and the 3D GPU. The latter is a complex discussion that involves cooperation from several parts of the industry, so we won’t get into that here.

As for the former, we’d certainly appreciate a bit more openness regarding image quality; anyone who ever bought a digital camera knows that endless lists of buzzwords do not always result in a pretty picture. Comparison pictures (‘before’, ‘after’, ‘competition’) would be a very good start, while ideally prototypes could be given early to digital camera review websites. Otherwise, how is this truly convergence?

Want to comment? Click here!