Vol 19, Issue 35

August 29, 2005

Low-Power Multicore H.264 Processor

By Kevin Krewell

The title is my idea for the perfect processor, circa 2005. Or at least one I expect I could get VCs to fund these days. The industry is buzzing with talk about power-efficient designs. Intel put "multicore" on our minds in 2004, when it switched the processors on its roadmap en masse to core scalability instead of clock-frequency scalability. Intel wasn't the first to talk multicore, but it talked the most about its benefits—and when Intel talks, the industry listens. And video processors have become a very trendy thing to build. Especially as the next wave of video-compression standards, based on MPEG-4, part 10 (http://en.wikipedia.org/wiki/H.264), is beginning to propagate throughout the industry. The new compression standards aim to upgrade the aging MPEG-2 standard with higher high-definition picture quality and lower bandwidth requirements. The cost of performing this algorithmic magic is increased design complexity and higher processing performance. A host of new companies, some led by well-known industry veterans, has jumped on the H.264 bandwagon.

This video-processor startup boom provides a strong sense of déjà vu. It reminds me of the network processor (NP) bonanza of the late 1990s, when a series of startups developed and marketed unique processor architectures designed to solve data-plane processing issues in a programmable way. The basic idea behind the NP startups was that the programmable solution could adapt to changing requirements and services, whereas the traditional ASIC solutions could not. These NP companies were mostly absorbed by larger companies like Agere (a spinoff of Lucent), Freescale (part of Motorola at the time), and Intel. Then there was a similar MPEG-2 bonanza, with a comparable explosion of innovative companies, like C-Cubed, Chromatics, and Zoran, building dedicated processors. The eventual result for the MPEG-2 processors was that they were superseded by faster processors, and by partial acceleration by graphics processors, and became irrelevant in PCs, although some found a market in embedded SoC implementations.

This is an example of the ongoing cycle in the industry, driven by people who want to build chips and by VCs looking to fund the next killer chip. Unfortunately, the VCs often wind up funding similar chips that are going after the same markets. The question for these H.264 vendors is how long the window for dedicated processors will remain open before the GPU vendors can integrate this function.

From left to right: Nick Tredennick, Brian Case, John Wharton, and Michael Slater, circa 1991, planning the Fourth Annual Microprocessor Forum. Source: MPR

Saying Good-bye to an Old Friend

One member of our editorial board who has seen a startup or two is Dr. Nick Tredennick. He's been founder of a few and adviser to many. Unfortunately, I have to report that Nick will be retiring from our board. His workload requires him to curtail his time on Microprocessor Report. He is a prized counselor to many startups, as well as the editor of the Gilder Technology Report.

In addition to being a member of the MPR editorial board, he has presented at every Microprocessor Forum since its inception (see photo). His "awards" at the Forum have provided us with humor in the midst of all the microarchitectural and product-roadmap seriousness.

Nick Tredennick is a funny guy, a smart guy, and an excellent writer—a dangerous combination. He's written for Microprocessor Report. He has voiced his support for configurable logic in the face of market apathy (see MPR 5/3/04-01, "Microprocessor Sunset"), because he believes configurable logic provides a better system solution with regard to power, chip development, and gate count than microprocessors do. Nick's dissection of Intel's dilemma with the Merced (Itanium) design (see MPR 9/13/1999, "Mercy, Mercy, Merced") was a classic. He has also authored a book (along with Brion Shimamoto): Microprocessor Logic Design (Bedford, MA: Digital Press, 1987).

Nick may be a bit embarrassed by my tribute to his contributions to this organization, but I think I can speak for all of us at Microprocessor Report in saying we'll miss his contributions to the report and his presence at the Forums.