[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
transmeta
i recieved a copy of an article on transmeta. makes for interesting reading.
maybe the shape of things to come?
--
Thanks,
Ross
ross.h [at] ntlworld.com
____________________________________________________________________________
{\rtf1\ansi\deff0
{\fonttbl{\f0\froman\fprq2 times;}{\f1\froman\fprq2\fcharset0 Times New Roman;}{\f2\fswiss\fprq2 helvetica;}{\f3\fnil\fprq2\fcharset0 Hightower-TextRoman;}{\f4\fswiss\fprq2\fcharset0 Arial;}}
{\colortbl\red0\green0\blue0;\red255\green255\blue255;\red128\green128\blue128;}
{\stylesheet{\s1\f1\fs20\snext1 Standard;}
{\s2\sb240\sa120\keepn\f2\fs28\sbasedon1\snext3 Heading;}
{\s3\f1\fs28\sbasedon1\snext3 Text body;}
{\s4\sa60\keepn\f3\b\sbasedon1\snext1{\*\soutlvl0} Heading 1;}
{\s5\li0\ri0\fi720\f1\fs28\sbasedon1\snext5 Hanging indent;}
{\s6\sa60\f4\fs20\b\sbasedon1\snext6 WW-Body Text 2;}
{\s7\sa60\f4\b\sbasedon1\snext7 WW-Body Text 3;}
{\s8\li0\ri0\fi720\f4\fs20\sbasedon1\snext8 WW-Body Text Indent 2;}
{\*\cs10 WW-Default Paragraph Font;}
}
{\info{\title On a few historic occasions, a high-tech company has placed an }{\author Justin}{\creatim\yr2000\mo1\dy20\hr20\min41}{\operator Red Herring}{\revtim\yr2000\mo1\dy20\hr21\min29}{\printim\yr2000\mo1\dy20\hr13\min29}{\comment StarWriter}{\vern5170}}\deftab720
{\*\pgdsctbl
{\pgdsc0\pgdscuse195\pgwsxn12240\pghsxn15840\margrsxn1800\margtsxn1440\margbsxn1440\pgdscnxt0 Standard;}}
\paperh15840\paperw12240\margl1800\margr1800\margt1440\margb1440\sectd\sbknone\pgwsxn12240\pghsxn15840\margrsxn1800\margtsxn1440\margbsxn1440\ftnbj\ftnstart1\ftnrstcont\ftnnar\aenddoc\aftnrstcont\aftnstart1\aftnnrlc
\pard\plain \s1\f1\fs20\li0\ri0\fi720\sa60\fs28 TRANSMETA\rquote S BIG GAMBLE
\par
\par The upstart chip maker bets it all on a mobile, post-PC era.
\par BY JUSTIN HIBBARD
\par
\par On a few historic occasions, a high-tech company has placed an enormous bet on a technology with the potential to transform an industry\emdash or fail spectacularly. In 1984, Apple Computer risked its future on the Macintosh and changed computing. Three years lat
er, GO Corporation tried to convert the world to pen computing and wound up losing $75 million. Now comes Silicon Valley\rquote s latest high-stakes gamble\emdash Transmeta Corporation.
\par \pard\plain \s5\li0\ri0\fi720\f1\fs28\sa60 Five years and more than $100 million in the making, Transmeta was conceived in the tradition of Silicon Valley\rquote s legendary firms. Like Intel, Apple, and Sun Microsystems, it aims to build not just a company but also a technology platform from which thous
ands of businesses might rise. \ldblquote With a company like this, you\rquote re trying to create a new industry,\rdblquote says William Tai, a Transmeta board member and a partner at Institutional Venture Partners.
\par Transmeta\rquote s product is a microprocessor, the brain that controls a computer. But this microprocessor is designed for a new era in which portable devices will supplant desktop PCs as the most common means of connecting to the Internet. These devices will gi
ve rise to Internet applications not limited by location or by the PC\rquote s deskbound, document-centric metaphors. Voice and data will be available everywhere, anytime, via handheld computers, cellular phones, and devices not yet invented, changing the way peo
ple live and work.
\par No one knows when\emdash or whether\emdash this era will arrive. But one factor that could help it arrive soon is a standard platform. The PC prevailed because Microsoft and Intel built a platform on which an army of PC makers and software developers built compatible pr
oducts. Transmeta proposes a standard platform on which mobile device makers and Internet software developers can build products for the post-PC era.
\par \pard\plain \s1\f1\fs20\li0\ri0\fi720\sa60\fs28 The company faces tough competition from other of chipmakers, especially Intel. When Transmeta started in 1995, Intel\rquote s mobile processors consistently lagged behind its desktop processors in performance, leaving it vulnerable to competition from mobile spe
cialists. Since then, Intel claims it has closed the gap between its mobile and desktop products. Last year, the company acquired StrongARM, a low-power chip for small devices. Now the burden is on Transmeta to differentiate its products from Intel\rquote s.
\par Transmeta\rquote s chief differentiator is its processor architecture. Unlike most processors, which consist almost entirely of silicon, Crusoe processors are half hardware, half software. By using software to do much of what chip hardware usually does, the Crus
oe processors are smaller and less expensive than conventional processors, and consume less power.
\par
\par MY CHIP FRIDAY
\par The new chips were named after Robinson Crusoe, the fictional adventurer who sailed the Pacific. \ldblquote Crusoe has a metaphor of mobility and travel,\rdblquote says James Chapman, Transmeta\rquote s vice president of marketing. But Crusoe also brings to mind a shipwreck. Daniel
Defoe\rquote s hero spent 28 years stranded on a desert island.
\par Long before Transmeta introduced its platform, the company attracted extraordinary attention, for three things: its star software engineer, Linus Torvalds, creator of the Linux operating system; high-flying investors like Paul Allen and George Soros; and s
ecrecy worthy of the C.I.A.
\par \pard\plain \s5\li0\ri0\fi720\f1\fs28\sa60 The company broke its five-year silence on January 19 when it introduced Crusoe in a historic villa in the foothills of the Santa Cruz Mountains. The headliner was David Ditzel, the gregarious co-founder and CEO with a knack for recruiting star engineers.
Mr. Torvalds made a cameo appearance, to demonstrate a Crusoe-powered machine by playing Quake against one of its creators, Dave Taylor, another Transmeta engineer. (Mr. Torvalds got slaughtered.)
\par \pard\plain \s3\f1\fs28\sa60 There are two Crusoe processors, one for notebooks, the other for \ldblquote Internet appliances,\rdblquote anything smaller than a notebook and larger than a palm-top. Transmeta claims both chips consume about one watt of electricity on average, allowing a noteboo
k to run all day on one battery. The appliance chip runs at 400 MHz, the notebook chip at 700 MHz. The notebook chip runs Windows applications about as fast as Intel\rquote s 500-MHz mobile Pentium III, Transmeta says.
\par \pard\plain \s1\f1\fs20\li0\ri0\fi720\sa60\fs28 At its product launch, Transmeta declined to name companies using its chips in their products. But executives said Crusoe-based products will be introduced throughout this year. The company has designed specifications for several products that use its chip
s, including a tablet computer that connects to the Internet via wireless technology.
\par During the event, Mr. Ditzel explained how Transmeta will ensure that software programs written for Intel chips run on Transmeta\rquote s chips\emdash an unofficial requirement to compete in the microprocessor business. Every chip has its own set of instructions that so
ftware programs use to make computers do things like save and print. Software companies must write different versions of their programs for each chip at considerable time and expense. Few of them will write programs for a new chip with little market share
and little hope of capturing a market that Intel owns. But thousands of companies have written programs for Intel\rquote s instruction set, called x86. To run those programs, other chips must support x86, too.
\par \pard\plain \s3\f1\fs28\sa60 The x-86 instruction set has grown complex over the years, requiring more transistors, which consume more power. Moreover, when chipmakers copy patented portions of Intel\rquote s chip design, they must pay licensing fees or battle Intel in court.
\par One of Transmeta\rquote s breakthroughs is what it calls \ldblquote code morphing software,\rdblquote which translates x-86 instructions into instructions that Crusoe chips can process. By using software, rather than hardware, to do the translation, Transmeta can build a sm
all, inexpensive, low-power chip that runs Windows applications \endash and avoid licensing fees and lawsuits. \ldblquote I\rquote m pretty sure that Transmeta is out of the woods on most of those patents,\rdblquote says Richard Belgard, an independent consultant who has studied Transmet
a\rquote s patent filings.
\par A software-based microprocessor is easier to repair or improve than a hardware-based one. Changing instructions in traditional chip hardware can take up to a year. Transmeta can change instructions in its software in a day and send the changes to a custome
r via the Internet. Or Transmeta can rapidly replace an instruction set. The company could even build several instruction sets into its software, allowing people to use applications for Windows, Macintosh, Unix, and Linux on one machine.
\par In theory, changes to Transmeta\rquote s software could improve the processor\rquote s performance. The company could sell the same chip hardware with two versions of software, one designed for greater speed than the other and each priced accordingly. To upgrade a proce
ssor, users could simply enter a credit card number on a Web site and download software. \ldblquote The marketing possibilities are astonishing,\rdblquote says George Sarlo, a partner at Walden Venture Capital, a Transmeta investor. Whether manufacturers who use Transmeta\rquote s
chip in their computers will want to control the software distribution remains to be seen.
\par
\par \pard\plain \s4\sa60\keepn\f3\b\li0\ri0\fi0\f1\fs28 CALCULATED RISC
\par \pard\plain \s6\sa60\f4\fs20\b\f1\fs28 Transmeta is not Mr. Ditzel\rquote s first effort to simplify microprocessors through the use of software. In fact, his efforts began 20 years ago. In 1978 he earned his master\rquote s degree in computer science from the University of California at Berkeley, where he s
tudied alongside future stars like Eric Schmidt and Bill Joy, cofounders of Sun Microsystems. In 1980, while working as a chip designer at AT&T\rquote s Bell Labs, he collaborated on a paper with one of his Berkeley professors, David Patterson. "The Case for the
Reduced Instruction Set Computer" proposed what was then a radical idea: engineers could make a processor run faster by reducing the number of instructions fed to it.
\par \pard\plain \s1\f1\fs20\sa60\fs28 An obscure technical newsletter published the paper, and The Reduced Instruction Set Computer (RISC) quickly won fans throughout the engineering community. But it was heretical to engineers who believed in a conventional approach called the Complex Instruc
tion Set Computer, or CISC, which uses elaborate instructions and requires complex circuitry. CISC devotees soon published a rebuttal, and what became known as the RISC-CISC debate began.
\par Within ten years, this esoteric dispute had escalated to a holy war that pitted multibillion-dollar corporations against one another, and divided the computer industry into two factions. By the 1990s, companies like Sun, Hewlett-Packard, IBM, Motorola, and
Apple had staked their businesses on RISC. Intel and Microsoft had built a PC empire based on CISC. The Macintosh versus the PC, Windows versus Unix, the Pentium processor versus the PowerPC processor\emdash all these rivalries stemmed from "The Case for the Red
uced Instruction Set Computer."
\par Mr. Ditzel proved to be not only a gifted scientist but also an effective spokesman for his ideas. In 1982 he organized one of the first conferences that featured a RISC-CISC debate. He persuaded IBM to let one of its researchers speak at the conference ab
out RISC technology the company had been quietly developing since 1976. Mr. Ditzel's gift for persuading other scientists to join his causes would become a hallmark of his career, as evidenced by the top-ranked engineers he attracted to Transmeta. "Dave is
the world's best recruiter," says Mr. Patterson, his co-author on the RISC paper.
\par While promoting RISC in the engineering world, Mr. Ditzel was helping design a RISC processor inside AT&T, called CRISP [what does CRISP stand for?-je]. [C-language Reduced Instruction Set Processor - JH] CRISP was never released, but the design became the
basis for the Hobbit, a low-power RISC chip that AT&T developed in 1988 for handheld computers. Foreshadowing Transmeta, AT&T was one of the first companies to see that it could exploit the simple circuitry of a RISC processor to produce a chip that consu
med little battery power.
\par \pard\plain \s7\sa60\f4\b\f1\fs28 In 1987 Mr. Ditzel joined Sun, where he worked for eight years as an architect and spokesman for scalable processor architecture (SPARC), the RISC design underlying Sun's processors. Eventually, he joined the growing number of engineers who believed microp
rocessors were becoming too complex. Even RISC processors, designed originally for simplicity, were gaining more and more transistors. Mr. Ditzel started thinking about how to simplify processors. He decided software was the answer.
\par \pard\plain \s1\f1\fs20\sa60\fs28
\par TRANSFIXED
\par Transmeta was incorporated on March 3, 1995, to bring forth Mr. Ditzel\rquote s vision of building software-based microprocessors. He went to work recruiting engineers who could pull off the technological feat. \ldblquote We realized there were less than a dozen people in
the world capable of doing this,\rdblquote Mr. Ditzel says. \ldblquote We had one simple plan: hire all of them.\rdblquote And Transmeta did\emdash not only to gain their expertise but also to keep them out of competitors\rquote ranks.
\par The first venture capital firm to invest in Transmeta was Walden Venture Capital. The firm saw an opportunity for Transmeta in the mobile computing market, where Intel seemed more vulnerable than in the desktop market. "You've got to have a niche where In
tel cannot compete, and that was the whole idea behind Transmeta," says George Sarlo, a partner at Walden. "It was physically impossible for Intel to have the power and the performance of Transmeta at the same time." Walden brought Institutional Venture Pa
rtners into the deal, and in December, 1995 both firms invested about $5.5 million in Transmeta.
\par By 1997, Transmeta had a commitment from IBM to manufacture its chips and provide technical support. The deal helped Transmeta attract a new cadre of prestigious investors. In April 1997 the company closed a $15 million round of financing that included par
ticipants like Vulcan Ventures (the VC operation of Microsoft cofounder Paul Allen), Integral Capital Partners, and, once again, Institutional Venture Partners and Walden, who also brought in Novus Ventures.
\par \pard\plain \s6\sa60\f4\fs20\b\f1\fs28 Also in 1997, Transmeta hired Linus Torvalds, the creator of Linux. At the time, Mr. Torvalds had several job offers, many of them from Linux companies. On his first visit to Transmeta, he toured the company for a day. \ldblquote I went back to my hotel in the eveni
ng, and I thought, these people are crazy,\rdblquote he says. \ldblquote When I came in the next day, I had decided that if I was going to work for a company, I wanted to work for a company that was doing something fun and interesting. I still haven\rquote t seen anything as exciti
ng as Transmeta.\rdblquote
\par \pard\plain \s6\sa60\f4\fs20\b{\fs28\f1{\b0 Though Mr. Torvalds was known as the ultimate authority on Linux, Transmeta hired him for his general software engineering skills. He worked on Transmeta\rquote s code-morphing software.} }
\par \pard\plain \s1\f1\fs20\sa60\fs28
\par SLOW METABOLISM
\par \pard\plain \s6\sa60\f4\fs20\b\f1\fs28 In January, 1998, Transmeta produced the first prototype of its processor at IBM Microelectronics\rquote foundry. The chip did not perform at the level the company had expected \endash not unusual for a prototype. According to a former Transmeta engineer, the company
thought its processor would run x86 applications at speeds comparable to a 400-MHz Pentium II. Instead, it ran them about as fast as a 200-MHz Pentium I. Other aspects of the chip met expectations \endash it consumed little power, and it translated instructions
accurately. Transmeta\rquote s engineers went to work fixing the bugs and improved the performance.
\par \pard\plain \s1\f1\fs20\sa60\fs28 Mr. Ditzel dismisses the incident as a mere test. But some insiders say it led Transmeta to refocus its target market. "There was a time when we thought the promise of the architecture was powerful enough that we could go after almost every market," says W
illiam Tai, a Transmeta board member and partner at Institutional Venture Partners. "But we realized we couldn't do everything. So we dug down in our roots and said, 'What are we really good at?'" Given Transmeta\rquote s low power consumption, the answer was mob
ile Internet devices.
\par To be sure, the company squeezed the best performance it could get out of its chips. But speed comes at the expense of low power, and Transmeta had to decide which feature to emphasize. It went with its strength \endash low power. The company still claims it chi
ps perform competitively. But Transmeta has called for a new methods of measuring the performance of mobile processors that give more weight to power consumption.
\par [cut? Mr. Ditzel places the company\rquote s decision to focus on mobile devices at an earlier date. In its infancy, Transmeta knew it would compete in the PC market, but it wasn\rquote t sure which segments. \ldblquote We had toyed with the idea of doing desktops,\rdblquote Mr. Ditzel s
ays. \ldblquote As we came to understand the technology, we realized mobile was best.\rdblquote A turning point, he says, came when the company discovered in 1997 that it could cut power consumption by as much as 50 percent by programming its processor to learn which parts o
f a program a person uses most and optimize those parts to run most efficiently. That discovery, Mr. Ditzel says, showed Transmeta where its strength were.]
\par
\par \pard\plain \s4\sa60\keepn\f3\b\li0\ri0\fi0\f1\fs28 METAMORPHOSIS
\par \pard\plain \s1\f1\fs20\li0\ri0\fi720\sa60\fs28 Transmeta\rquote s strength happens to coincide with a market poised to boom. The number of mobile devices sold worldwide will jump from 3.9 million in 1998 to 21.3 million in 2003, according to market research firm Dataquest. That\rquote s still fewer than the 199 mill
ion PCs that Dataquest predicts will sell worldwide in 2003. But sales of mobile devices are growing at a compound annual rate of 45 percent, while PCs sales are growing at 22 percent.
\par \pard\plain \s1\f1\fs20\sa60\fs28 \ldblquote Targeting mobile devices and information appliances, which have tremendous upside potential, is smart,\rdblquote says David Hayden, an analyst at Mobile Insights. \ldblquote Transmeta has recognized that trying to duke it out in the desktop space is a losing proposition.\rdblquote
\par \pard\plain \s3\f1\fs28\sa60 Intel, king of the desktop, has its eye on the mobile market, too. Unlike five years ago when Transmeta was founded, Intel\rquote s mobile processors no longer trail its desktop processors in performance. In January, Intel introduced its 650-MHz mobile Pentium II
I processor with new power management technology. It was the third increase in mobile processor speed Intel had introduced in a year, nearly doubling the speed it offered a year earlier.
\par \pard\plain \s1\f1\fs20\sa60\fs28 Speed continues to drive sales of PCs. But in a new world of mobile Internet devices, there\rquote s no guarantee users will value performance over other criteria, like price and battery life. \ldblquote People buy laptops with a very clear focus on the clock speed,\rdblquote says
Joseph Osha, an analyst at Merrill Lynch. \ldblquote But people don\rquote t do that with portable devices. In a new market, people don\rquote t care what kind of chips are in the machines.\rdblquote
\par In the sort term, that may be good for Transmeta. It means there is an opportunity for an upstart chipmaker to distribute its chips widely in a new class of devices. But eventually, it will serve Transmeta well if people care whether a Crusoe processor is
in their device. And it will serve the burgeoning mobile Internet device industry to have a standard platform on which devices and applications are based.
\par A standard mobile Internet platform will promote the development of Internet applications that go beyond PC functionality. Current Internet applications are based on features used in PCs more than 20 years ago. Web pages, for example, are relics of the doc
ument metaphor used in early PC graphical user interfaces in the 1970s. When the Internet is available through small devices that can roam anywhere, applications will emerge that we haven\rquote t yet imagined.
\par Even if Transmeta doesn\rquote t bring these applications into the world, it shows us that they are possible.
\par
\par \pard\plain \s1\f1\fs20\sa60\fs28\i Red Herring Online reporters Larry Aragon and Phil Harvey and Red Herring research associate Tom Geck contributed to this story.
\par }