|The Transmeta mystery|
|Written by Sean M. DuganHits : 478|
|Wednesday, 31 January 2001 19:00|
Is the hot new crusoe chip a wonder or a blunder?
Sometimes a whisper gets more attention than a shout. For nearly five years, Transmeta, a startup chip maker based in Santa Clara, Calif., was a conundrum that would have baffled the Great Detective. Cloaked in a veil of secrecy, the processor maker attracted millions of dollars in investments from the likes of AOL, Compaq, Gateway, Samsung, and Sony. Transmeta assembled an all-star cast of technical talent, including geek superstar Linus Torvalds, the creator of the Linux operating-system kernel that bears his name. Only a few patent filings in 1999 hinted as to the company's ultimate goal--something to do with chips and software emulation. Then, in January 2000, amid the sort of media attention usually reserved for long-awaited albums from top rock bands, Transmeta unveiled its hidden treasure: a power-saving microprocessor that promised to usher in a new era of mobile computing and threaten Intel's crown jewels.
A year later, much of the mystery swirling around Transmeta remains. Is its Crusoe chip for real? Is the company destined to become a powerhouse in ultralight computers and Internet appliances, or a footnote in chip history? In the past few months the company has suffered a number of setbacks, including lackluster scores on performance tests and rejection by several prominent computer manufacturers.
After widespread reports that IBM's new Thinkpad 240 would incorporate Crusoe, IBM announced last November--just days before Transmeta's IPO--that it wouldn't be using the chip after all. Compaq also said that it had no immediate plans for Crusoe in its line of laptops and notebooks. The biggest blow for Transmeta came last fall when NEC recalled about 300 notebooks because of apparent flaws in the Crusoe chip.
With analysts raising questions about Transmeta's technology, and Intel coming on strong in the mobile market with a new line of power-conserving Pentiums, 2001 is a pivotal year for the upstart company. Transmeta's carefully constructed curtain will be pulled back, revealing either a viable contender in the featherweight-chip wars or a pretender that succeeded only in spending other people's money.
Crusoe, the secret heart of Transmeta
In the hot-and-heavy world of microprocessors, there's a constant race to jam more transistors onto smaller and smaller wafers of silicon. More transistors means more processing power and faster computers.
But the chip-fabrication industry faces twin challenges: First, as transistors shrink, development costs grow exponentially. Second, the mobile-computer market presents different challenges than its desktop cousin; cooling is becoming a hot design problem. Hotter still is battery life. As processors continue to pack more power, they also eat more power. Up to 35 percent of laptop battery life is eaten up by the processor. So as laptops gain Hertz, battery life tends to shrink. Flat-panel screens, hard drives, and accessories have not effectively reduced power drains. And battery technology is not keeping up with demand. So it behooves the mobile-processor maker to fabricate a low-power chip.
Couple this with the fact that users expect their laptops to perform at desktop levels and the magnitude of these design problems begins to surface. Intel's solution is to give users what seems like the best of both worlds: Run the laptop at full speed while plugged in and at special low-power settings when on batteries. Software allows users to override the automatic power-down if full power is needed on batteries, but users can expect no more than two hours of battery life under those circumstances. The downsides include a need to work with OEMs on cooling issues and the aforementioned cost issues.
The Crusoe processor, the brainchild of Transmeta's CEO David Ditzel, takes a different approach: It considered power, price, and heat dissipation as the core design problems. The heart of Crusoe is what Transmeta calls code morphing, a technology that carries out certain chip-level instructions with software rather than hardware components. Transmeta claims that this unusual architecture makes its chips cheaper--they're not as complex, and require fewer transistors than Intel's Pentium family--and easier to develop and improve on. It's comparatively easy to fine-tune Crusoe's software layer, rather than redesigning the chip and having to produce new chips. "Our technology is fundamentally different (from Intel's)," says Ed McKernan, Transmeta's director of marketing.
To duplicate the functionality of Intel's Pentium chips, Crusoe's software emulates the x86 instruction set that drives the Windows and Linux operating systems. But emulation traditionally hurts performance, running programs slower than a chip that natively executes instructions. The ace up Transmeta's sleeve is dynamic optimization, a system that allows the chip to reprogram itself on the fly to deliver optimal performance. In theory, Transmeta's code-morphing technology could meet or even exceed the performance of Intel chips, because of the tangled legacy of the x86 instruction set.
Smoke and mirrors?
Translating theory into reality has been problematic for Transmeta, however. Uniting the largely separate worlds of hardware and software engineering is difficult, even for Torvalds and Co.
Since the debut of Crusoe, Transmeta has faced challenging questions about how well its Crusoe performed in comparison with Intel and AMD chips. In benchmarks tests by ZDNet and PC Week, Crusoe lagged behind its competitors.
Transmeta has defended itself by stressing that traditional benchmarks aren't a fair test of Crusoe's potential, because of its ability to adapt to a user's needs, performing repeated tasks more efficiently. McKernan says that battery life is more important than sheer clock speed in the nascent markets that Transmeta is going after--Internet appliances and computers weighing less than four pounds. He says that the company plans to release studies this year showing productivity gains from using Crusoe-powered devices. "Crusoe will be bought for productivity, not performance," McKernan says. "The productivity of a dead battery is zero."
Indeed, according to research conducted by International Data Corp., processor speed is not the most important criterion by which users judge subnotebooks; battery life and lightness both rank ahead of overall system performance. David Clingerman, a software engineer at Sun Microsystems who owns an HP Jornada 690 handheld PC, agrees with that assessment. "For me, it is size, battery life, and performance, in that order," he says. "I would certainly like more power in my HPC--as long as it does not come at the expense of form factor or too little battery life."
By extending cordless work time, Transmeta has "developed an important new design point" for small, sub-3-pound notebooks and tablets, says Carl Howe, an analyst with Forrester Research. "Combining all the attributes you want in this market, Transmeta is the only processor."
Adds McKernan: "In the end, the OEMs know the difference."
The question is, do OEMs know the difference? Sure, some subnotebook and PDA manufacturers seem to have embraced Crusoe as an alternative to chips from Intel, AMD, Cyrix, or National Semiconductor. Sony has recently released the Vaio PictureBook, a 2.2-pound subnotebook with an integrated digital camera. Crusoe's power savings made it "the clear choice" for a device aimed at the highly mobile user, says Mark Hanson, general manager of the Vaio PC line. "We got times-two battery capacity compared to previous offerings." Meanwhile, Gateway has teamed up with AOL to offer the Connected Web Pad, an inexpensive, Crusoe-powered Internet appliance with a touch screen, and Hitachi has launched its Floria 55mi, a combination cellular phone and handheld PC. Most of Transmeta's current customer base is in Asia, where computing devices are a fashion accessory and subnotebooks are extremely popular. "In Japan, an 8-pound Dell notebook looks as out of place as a Cadillac," Howe says. The Floria is currently available only in Japan, but Hitachi is considering it for the American market.
But other OEMs have backed away from Crusoe or demonstrated their disappointment in the chip. Notably absent from Transmeta's presentation at Comdex last month was any mention of IBM or Compaq. IBM spokesman Raymond Gorman explains that Crusoe just didn't measure up to the Pentium III as a power plant for the company's new notebook.
"The processor is only part of the story of extended battery life," Gorman says. "Many other components--including the screen, software, and chipset--help determine battery life. As the integrator of these technologies, IBM must carefully consider all aspects of the ThinkPad system design to balance battery life with performance. After exhaustive testing and evaluation, we didn't believe there was a significant advantage to using Transmeta Crusoe-based chips in the Thinkpad 240."
That decision probably dampened enthusiasm for Transmeta's IPO, which nevertheless was fairly successful, debuting at $21 a share and soaring 115 percent on the first day.
NEC's recall was the biggest blow to Transmeta's reputation and mystique. Declining to cite a specific number, McKernan said that a "very small amount" of Crusoe processors had flaws. While flawed computer chips are nothing new--Intel suffered through several Pentium recalls in the '90s--the problems do heighten concern about the unproven company. "After the latest recall, vendors will think before they switch to a new platform," says Josephine Mong, a senior research analyst at IDC.
Will the real Transmeta please stand up?
Mong and other analysts question whether Intel is really Transmeta's chief enemy. "Transmeta is too focused on competing with Intel," she says. "Their real competition is companies like National Semiconductor with its Geoprocessor." Whether Transmeta is gunning for Intel or not, it has spurred Intel to action, making the king of silicon rethink its strategy. "Extended battery life is increasingly important to today's mobile-computer user, and Transmeta woke up the industry to the importance of battery life in notebooks," IBM's Gorman says.
In the past year, Intel has introduced Pentium and Celeron chips that consume much less power than their predecessors, rivaling Crusoe's thriftiness. "Intel is not going to let a small company like Transmeta take away its market," Mong says.
To survive in an extremely competitive industry, Transmeta must prove that Crusoe is for real, and achieve much deeper market penetration in North America. McKernan says that Transmeta is already addressing the performance issue, trying to boost both processing speed and battery life. "In Q1, you'll see an enhancement to Crusoe that will improve it 20 percent across the board," he says. Thereafter, the company plans to release a new version of Crusoe every six months, offering a 20 percent boost in performance with each upgrade. And McKernan sees a burgeoning market in the United States and Canada for Transmeta, encouraged by statistics that indicate that the tide is shifting toward ultraportables and simple appliances. IDC expects the market for Internet appliances to jump to $17.8 billion in 2004, up from a mere $2.4 billion in 1999. In an IDC survey last year, 41.5 percent of U.S. companies said they planned to purchase ultraportables. This year, some of those machines may even be manufactured by NEC, which is apparently willing to give Transmeta a second chance. Beth Makosy of NEC says that the Japanese firm will be evaluating Crusoe for North America at the time this story is in print.
Transmeta has sparked the imagination of the chip industry, as well as igniting passionate debate about the direction of personal computing. Whether Crusoe inspires the development of a new breed of miniature computers or ends up on the bottom shelves of used-computer stores remains to be seen. Howe says it's simply too early to tell if Transmeta will become a powerful force in the industry. McKernan insists that Transmeta is firmly committed to its mission to produce chips uniquely suited to mobile computing. "Our strategy hasn't changed since January ," he says. "The sub-4-pound ultralight market is the future of mobile computing."
Sean M. Dugan is a contributing editor and columnist for ComputerUser and InfoWorld.