January 30, 2006

Editor: Tom R. Halfhill

In this issue:

• Cell Processor Isn’t Just for Games

• Embedded Processors Thrive in 2005

• MPR’s Best Server of 2005

• MPR’s Best PC Processor of 2005

• Best DSP for Audio and More

• Freescale’s MXC Voted Best

• Cortex-A8 Balances Power, Performance

• This Technology Is Virtually Here Now

• Editorial: A Look Ahead to 2006

Cell Processor Isn’t Just for Games

Deciding on our MPR Analysts’ Choice Award for Best High-Performance Embedded Processor of 2005 wasn’t easy. We evaluated several strong candidates before picking our winner: the Cell Broadband Engine, jointly designed by the STI alliance: Sony, Toshiba, and IBM Microelectronics.

The Cell BE is destined for Sony’s next-generation home videogame console, the PlayStation 3, scheduled for release later this year. However, unlike the processors that IBM designed for Sony’s competitors—unnamed CPUs buried inside the Microsoft Xbox 360 and Nintendo Revolution—the Cell BE is part of a grander scheme. Sony, Toshiba, and especially IBM are promoting this radical multicore processor and future derivatives as popular number-crunching engines for everything from cellphones to supercomputers. (See MPR 2/14/05, “Cell Moves Into the Limelight,” and MPR 1/3/05, “New Patent Reveals Cell Secrets.”)

Their vision may seem overly optimistic, as we noted in an editorial last year. (See MPR 2/28/05, “Cutting Through Cell’s Hype.”) But it’s not just a dog-and-pony show. Toshiba, a manufacturing partner in the Cell alliance, plans to use the chips in high-volume consumer-electronics products, such as high-definition TVs. Sony has similar plans.

Outside the STI alliance, three customers have adopted the Cell BE: Mercury Computer Systems, Raytheon, and Stanford University. Mercury is showing three products: a blade server, a rugged “computing appliance” for military systems, and a powerful workstation for scientific, technical, and industrial applications. Raytheon says it will use the Cell BE in missile systems, artillery shells, and radars. Stanford University is building a supercomputer for scientific computing. These impressive design wins prove that the Cell BE isn’t only for fun and games.

Microprocessor Report readers can access the full story (2 pages) here: www.mdronline.com/mpr/h/2006/0130/200501.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Embedded Processors Thrive in 2005

Looking back at 2005, we can identify five broad trends in embedded processors. None of these trends actually started last year, but they gained momentum and will be major forces in 2006 and the years to come.

• The multicore revolution that broke into the mainstream in 2004 is accelerating. High-performance embedded processors with half a dozen or more processor cores are becoming almost commonplace. Now the attention is expanding to software—particularly the need to create new development tools and techniques for exploiting parallelism.

• Startups and new microprocessor architectures keep emerging, undeterred by entrenched competition. Inventors have no shortage of fresh ideas. Fortunately, the investment community has largely recovered from the shock of the 2000 tech bust and is willing to fund new ventures with serious money. Venture capitalists are again prowling our Microprocessor Forums.

• More alternatives to designing custom chips keep surfacing, driven by the soaring costs of deep-submicron mask sets and nonrecurring engineering (NRE). A key ingredient in many alternatives is programmable logic. FPGA prices keep falling, steadily altering the equation that determines when programmable-logic devices become more economical than custom ASICs and SoCs. An equally important development is new types of reconfigurable processors that combine programmable logic with fixed logic.

• Markets attracting the most attention from chip designers and embedded-system developers are mobile communications, networking, electronic games, digital audio, and digital video. Moreover, the convergence of consumer electronics long anticipated for the living room is now happening in pockets as well.

Silicon Valley is gradually losing its status as the unofficial capital of the semiconductor industry. The industry is becoming much more diffuse, due mainly to unprecedented growth in the Asia/Pacific region and India. Companies lacking a multinational strategy or a widely dispersed workforce are often regarded as too small or old-fashioned to compete in world markets. Outsourced engineering is routine. China is becoming more assertive and is destined to become both a world standard-setter and a major developer and exporter of embedded processors.

Microprocessor Report readers can access the full story (9 pages; 6 graphics) here: www.mdronline.com/mpr/h/2006/0130/200502.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

MPR’s Best Server of 2005

The MPR Analysts’ Choice for Best Server Processor of 2005 is the Dual-Core AMD Opteron processor. The competition was tough this year, as it has been in any year we’ve presented this award.

Fred Weber’s presentation at Microprocessor Forum 2001 revealed that the Opteron design was always meant to scale to a dual-core version. In contrast, Intel’s Paxton XeonMP was clearly rushed to market as two cores slammed onto one die. But Intel engineers did an admirable job by pulling the Paxton dual-core version of the Xeon processor into 4Q05 to meet the competition of AMD’s dual-core Opteron.

The competitors for best server that we considered this year included the perennial example of superior wide-execution RISC processors, the Power5 processor. The Power5 is undergoing a shrink from 130nm to 90nm (Power5+), but the only commercially available system we were familiar with at the end of 2005 is running the specifications of the original 130nm Power5, introduced in 2004.

A malaise befell the Itanium family as the Montecito version got pushed out to 2006 and likely lost the Foxton power-management feature touted at the past year’s Hot Chips symposium.

The most radical innovation came from Sun’s UltraSPARC T1 processor (Niagara), with eight simplified SPARC v9 cores on one die. The T1 delivered exceptional performance per watt on specific workloads, but the lack of significant floating-point performance and a coherent multichip interface limits its range of uses.

The cream of server manufacturers—such as Cray, HP, IBM, and Sun Microsystems—has built outstanding Opteron servers. The only major server vendor that hasn’t offered Opteron systems has been Dell Computers, which has stuck to its Intel-only strategy to date, although we do not believe this Dell policy will last much longer. The balanced performance, power, and pricing of the AMD dual-core Opteron processor have led us to decide it is the best overall server processor for 2005.

Microprocessor Report readers can access the full story (2 pages) here: www.mdronline.com/mpr/h/2006/0130/200503.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

MPR’s Best PC Processor of 2005

The design criteria for PC processors changed dramatically in 2005 to focus on increasing performance through the use of multiple cores, while reducing or maintaining the power and thermal envelopes of single-core processors. AMD, IBM, and Intel all introduced new dual-core processors for the desktop PC segment in 2005, with the Athlon 64 X2 taking top performer honors. The sole dual-core mobile processor from Intel, however, has proven to be the most innovative dual-core PC processor and a leading indicator of architectural design trends for future PC processors.

Microprocessor Report readers can access the full story (2 pages; 2 graphics) here: www.mdronline.com/mpr/h/2006/0130/200504.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Best DSP for Audio and More

Choosing the best DSP produced in 2005 might have been more straightforward if one had to consider only its signal-processing functions. There are, however, only a handful of chips left that can still be seen as “pure DSP” devices; most pure DSPs have become cores incorporated into complex chips along with fixed and writable memory, hard-wired accelerators, microcontrollers, and auxiliary DMA units.

Audio was undoubtedly the past year’s most targeted market in which most DSP chips and cores competed. MPR’s analysts therefore decided to recognize the best DSP-based audio chip for 2005, even as 2006 seems to be shaping up as the year of competition for video sockets.

Microprocessor Report readers can access the full story (2 pages; 1 graphic) here: www.mdronline.com/mpr/h/2006/0130/200505.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Freescale’s MXC Voted Best

In 2003, Freescale (then still operating under the Motorola flag) introduced the Mobile Extreme Convergence (MXC) Architecture, code-named Jupiter, at In-Stat’s Microprocessor Forum (MPF2003). Conceived during a period when many processor architects were employing heterogeneous architectures, Freescale’s daring brainchild took the opposite direction: Jupiter’s design converged two processor cores, a DSP and an MCU, into one engine plus a separate ARM-based applications processor.

Microprocessor Report readers can access the full story (3 pages; 1 graphic) here: www.mdronline.com/mpr/h/2006/0130/200506.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Cortex-A8 Balances Power, Performance

Years from now, the industry may remember 2005 as the pivotal year when ARM began extending its reach from low power to high performance. In any event, we believe ARM’s fastest processor to date—the Cortex-A8—deserves our MPR Analysts’ Choice Award for Best Processor-IP Core of 2005.

The Cortex-A8 is ARM’s first superscalar processor core, and it’s the first ARM processor capable of attaining clock frequencies in the gigahertz range. It’s the biggest departure in processor design for ARM since the company was founded in 1990, after spinning off from Acorn Computer.

For 15 years, ARM has specialized in licensing small, relatively simple embedded-processor cores that emphasize low power consumption over high throughput. It’s a successful formula that has made ARM the industry leader in licensable intellectual-property (IP) processor cores for chip integration. Although low-power processors remain ARM’s strength, the Cortex-A8 branches out in a bold new direction.

Microprocessor Report readers can access the full story (2 pages) here: www.mdronline.com/mpr/h/2006/0130/200507.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

This Technology Is Virtually Here Now

Although for 2005’s technology of the year award, we looked at a number of possibilities, we considered the breakout technology of 2005 to be the various implementations of computer virtualization. Virtualization in computing has many different instantiations and is related to emulation as well. But the newest move for virtualization is that it is now poised to change client computing. While virtualization has been around for a long time, mostly in servers, and the exact origin and adoption of virtualization would make a great historical article, a confluence of events began taking shape in 2005. These events included Apple Computer’s transition from PowerPC to Intel x86 processors; the efforts of Sony and Microsoft to add backward compatibility to their newest game consoles; the positive reception of XenSource’s presentation at Fall Processor Forum 2005; and the AMD and Intel development of x86 virtualization support in upcoming hardware.

Virtualization is a technology that can accelerate the move to multicore processors by creating an environment that can wring the most performance and the greatest cost advantage out of multicore technology. Virtualization is breaking the locks on architecture changes, allowing system developers to have greater choice in the future. Virtualization can also simplify the world of software deployment, where old software need not be updated for a platform transition. For consumers, it helps maintain the value of content they already own and can lead to more-robust PC platforms. Virtualization is a technology set to remake the world of computers for the better.

Microprocessor Report readers can access the full story (2 pages) here: www.mdronline.com/mpr/h/2006/0130/200508.html. To find out more about Microprocessor Report, please visit: www.mdronline.com.

Editorial: A Look Ahead to 2006

While this issue looks back at the best of 2005, I’d like to take this opportunity to look ahead to 2006 with a few modest predictions and requests. A caveat first: The problem with predicting the future without having any actual control over the people and companies involved leaves the predictions mostly guesswork and wishful thinking. I can make my predictions, but I cannot directly make those predictions come true. Other people must make these predictions come true through execution and decisions. So if I’m wrong, it’s someone else’s fault!

Now that I’ve completely disclaimed any responsibility for them, here are my predictions for 2006:

Apple Computer will quietly begin to talk to AMD about supplying processors sometime in the future. Not that Apple will actively try to undercut its new relationship with Intel in 2006, but it would be prudent to begin preparing for the future. Eventually, Apple will piss off Intel, as it has previous processor suppliers (Freescale and IBM), and will need a backup supplier. Apple will probably add AMD to its processor supplier list in 2008.

AMD will continue to gain significant market share in servers. It has always been clear to us that Opteron has been the more modern and efficient design for mainstream servers, compared with the hyper-pipelined NetBurst Xeon. Opteron was designed as a server processor that can be used for desktop and mobile, whereas the hyper-pipelined NetBurst microarchitecture was designed for media processing, not data processing. Overall AMD market share in client and mobile processors also looks to be gaining, but not as dramatically as in servers.

Editorial board member Fred Weber agrees and adds his own prediction that AMD will gain significant market share in commercial clients (desktop and mobile). Fred sees significant gains for Sun Microsystems, armed with Opteron servers and the new Niagara chip. Although Fred may be a bit biased toward AMD and AMD’s partner Sun, he is also quite bullish on the new Intel-based Apple computers and expects Apple to continue gaining mind share and market share, expanding its market to sweep up more entertainment devices.

In his CES keynote, Bill Gates declared that mainstream high-value, legal high-definition video content will come to media-center PCs later this year, but I also expect there will be many restrictions to that access. To get the digital high-definition content from studios and cable companies (as well as from satellite and phone companies), the high-value content will have to be highly protected from casual (and not-so-casual) piracy. Those restrictions will likely require the Intel Viiv platform, with Microsoft’s Vista operating system, and video connectors with HDCP (High-bandwidth Digital Content Protection) encryption on an HDMI (High-Definition Multi-media Interface) connection. This means you’ll probably need a brand new high-definition media-center PC and will be unable to upgrade an existing MCPC to high definition. I also expect that the Linux operating system will not qualify, unless it’s a highly customized version (like TiVO). The one wild card will be if Steve Jobs and Apple build Apple’s own version of a high-definition media PC.

The most obvious prediction is that there will be more media processors in 2006 focused on the H.264 standard. I expect there will be a shakeout and consolidation of those offerings in 2007.

What Won’t Happen in 2006
While there are often many predictions of what will happen in 2006, I have a few predictions of what I think will not happen.

The first prediction of 2006 nonevents is that despite the continued rumors of Itanium’s demise, neither Intel nor HP will back away from Itanium in 2006. Although the prospects for Itanium are not particularly bright, I think the Itanium program will continue to limp along for another few years. Fundamentally, the architecture concepts behind Itanium represent the thinking of the 1980s and ’90s wide-issue statically compiled code. Server workloads are moving toward virtualized, dynamically recompiled, highly threaded code, where Itanium has no strengths. Itanium also hasn’t blown its primary competitor, the IBM Power processor family, out of the water. If Intel had sunk the same resources into a more traditional RISC architecture (such as Alpha, which appears to be everyone’s favorite canceled chip) that it did into EPIC, we would likely be in roughly the same place—except that the Alpha architecture wouldn’t have been reviled as much by the competition as the VLIW-like EPIC architecture has been.

AMD will not pass 25% market share of the worldwide PC processor market in 2006. In fact, I don’t think AMD will ever pass 25% of the PC processor market. First, it would strain AMD’s fab capacity to supply the 60 million or more processors without reducing the average die size significantly and in this era of dual-core processors, the die sizes will actually have to grow somewhat before process shrinks can get them back under control. I also have a deep-seated belief that if AMD were to reach 25% market share, it would spark a price war with Intel that AMD lacks the financial strength to win. I saw it before, when I was at AMD. When everyone at AMD was convinced that Intel wouldn’t cut prices to meet AMD’s prices, because Intel would lose more money than AMD would, Intel did go for the jugular to deprive AMD of the revenue and market share opportunities. Never, ever, underestimate how strongly Intel will protect its position.

In a related prediction, I don’t think Dell will use AMD desktop and mobile processors in 2006. Dell is likely waiting to see how well Intel’s next-generation microarchitecture performs against AMD’s Hammer architecture. Dell will also likely wait to see how AMD’s Fab 36 and foundry suppliers ramp up in 2006. Finally, Dell will want to see how Intel treats Intel’s new best friend, Apple Computer, in 2006 before deciding if it (Dell) needs to teach Intel a lesson. I should note that, at the recent CES, Michael Dell was saying positive things about AMD (again). I do believe think that Dell will use AMD’s Opteron in 2007.

In 2006, John Dvorak will not get any less cranky.

Fun With Podcasts and Blogs
There is this new phenomenon where it seems that everyone wants to be a radio star. It’s called podcasting, named by combining the name of the most popular portable music player, the Apple iPod, with the reach of “casting” (as in broadcasting) the content out for many to download. Recently, Apple’s iTunes added a podcast section, making it extremely simple to find popular podcasts, download them, and transfer them to your iPod. I have become one of the enthusiasts of podcasts. In particular, I enjoy listening to This Week in Technology (also known as TWiT) with former members of TechTV, including host Leo Laporte and resident curmudgeon John Dvorak. The advantage of podcast is that I can listen to them whenever I want—in the car while commuting, in the gym while working out, or on my PC in the office. The only thing I find frustrating is that because the podcast is prerecorded, I find myself complaining to the iPod when a speaker gets something wrong.

I also like to listen to a locally produced public broadcasting show called GeekSpeak, on which I appeared earlier in 2005, and I’m exploring new shows about security and digital photography. Reed Business (our parent company) is exploring the use of podcasts with other tech publications, so we may wind up contributing to corporate efforts. I should note that not everyone wants to be a podcast star. Tom Halfhill recently mentioned that he specifically went into written journalism because he didn’t want to go into radio or TV!

The other movement that has been sweeping the media business is blogs (short for web logs). Blogs predate podcasts by a few years, and I have dabbled in my own blog in the past. I haven’t kept it up because I like to think my editorial page in MPR is my (very well edited) blog, even if it’s updated only once a month, I have doubts that readers will be fascinated by my daily thoughts. Sometimes blogging strike me as part narcissism (everyone will be thrilled to read my every thought); part marketing (the more I write about my company, the more people will believe we’re great); part popularity contest (my blog is more interesting than your blog; therefore I’m better than you). But often blogging can be a positive outlet for people to share ideas. I think that providing a mechanism for more direct and honest communication is a good thing.

What I don’t fully trust are corporate blogs. Because of issues of corporate liability, some editorial control must be considered by the blogger. My own experience with a similar situation occurred years ago when I had posted on an investor discussion board to correct someone on a technical detail, just as I did with engineers every day as an FAE. Later, I was chastised for posting information in a venue where investors communicated—that was the job of investor relations. Of course, I posted only factual information on an existing product, but it goes to show how nervous companies are about legal protections. (Another employee had posted reasonably confidential information on that same site and got into much bigger trouble.) Corporations will have to be mindful of investor lawsuits and potential loose-cannon employees. And everyone knows that posting negative information about your employer will get you fired.

Analysts and journalists also have a public image to protect (and project), so they will often self-edit comments in blogs, as well. Only people with little to lose, and often little to share, can be truly honest. Which is a long-winded way of saying that I don’t read many blogs. The only ones I check regularly are those I would call “professional” blogs. My favorites are those focused on the latest consumer devices: Engadget.com and Gizmodo.com; they are good for tracking the latest consumer devices and rumors. For example, during the 2006 CES, Engadget was posting a near-real-time blow-by-blow description of the keynotes. Our corporate cousin, EDN (ww.edn.com) also has a number of worthwhile blogs written by its editorial staff, and recently EDN, eNews, and Electronic Business posted a combined CES show site that included blogs.

To find out more about Microprocessor Report, please visit: www.mdronline.com.

• Most Recent Processor Watch Articles
• Processor Watch Archives