The Transmeta enigma

At a tantalizingly elusive Silicon Valley start-up, secrecy spawns hopes of revolution.

Published May 22, 1998 7:00PM (EDT)

The address for Transmeta -- a mysterious, supersecretive Silicon Valley start-up rumored to be working on a revolutionary new chip -- is 3940 Freedom Circle in Santa Clara. But according to my 7-year-old map of San Jose, Freedom Circle does not exist.

That, in itself, did not alarm me. Yesterday's maps are always out-of-date in the Valley. It's not unusual for high-tech business parks to spring up, fully formed, between one morning and the next -- especially here, practically under the shadow of Intel headquarters. If Silicon Valley had a ground zero, this would be it.

But my suspicions grew after I arrived at the doors of Transmeta, a one-story, low-slung stucco-roofed office building with impenetrably dark floor-to-ceiling windows. The trees dotting the parking lot looked much older than my dogeared map. Had the building been constructed around the trees? Or had these trees been shipped in, fully grown -- a minor detail of state-of-the-art landscaping practice? I'd seen it happen before.

As I wandered around the Transmeta building sizing up the trees, I reflected that I was acting just a bit paranoid. But with good reason, I reassured myself, as I glanced over my shoulder, half-expecting to see Thomas Pynchon lurking about. Think about it: The very name Transmeta -- a combination of Latin and Greek words that together could mean "above the beyond" or "across the next level " -- connotes a meaninglessness so vast it might be profound. Or maybe not.

Once upon a time, only a Pynchon would have dared such silliness. And only a Pynchon could have conjured up the Transmeta scenario: A start-up company backed in part by Microsoft co-founder Paul Allen attracts international attention by hiring one of the most famous programmers in the world -- free-software hero Linus Torvalds, creator of the Linux operating system. Then it refuses to say a single word about what he or the rest of the company is cooking up behind closed doors.

The story of Transmeta is laced with the kind of satirical geek humor that was once safely confined between the covers of novels by Pynchon or Neal Stephenson but is now irresistibly infecting the real-life operating system of Silicon Valley. The Web page for Transmeta says only: "This web page is not here yet." If you peek at the source code for the page, hoping for illumination, you're informed, "There are no secret messages in the source code to this Web page." In a particularly Pynchonesque stroke of ironic bravado, Transmeta even paid good money two years ago to a San Mateo consultant for the construction of a "corporate identity package." Apparently it's not easy, in the late '90s, to be a cipher. Even secrecy needs a branding campaign.

So what is Transmeta -- a company that makes the words "low profile" seem brassily exhibitionist -- up to? The consensus is that the hardware engineers and software programmers at Transmeta are "top-notch" and "incredibly bright." "They've got an unbelievably dense crop of talent," says EE Times technology columnist Alexander Wolfe. But no one seems to know what all these brilliant minds are doing. All roads to Transmeta lead straight to zipped lips sealed by nondisclosure agreements.

"They have been incredibly secretive," says Nathan Brookwood, chief chip analyst for Dataquest. "They're not talking, haven't been talking, and they've been at it for over a year, almost two years. You'd think they would have something by now."

"I've known [Transmeta CEO and founder] Dave Ditzel for 15 years," says independent chip designer John Wharton, "and I still don't have the foggiest notion what they are doing. I've asked Dave, and he just smiles and says, come on down, sign an NDA and I'll tell you."

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - -


Despite the fact that hardly anybody who hasn't signed an NDA can say an intelligent word about Transmeta, many technology watchers on the Net believe the company is up to "something wonderful" -- something that will change the computing world.

After all, Transmeta hired Linus Torvalds. Torvalds is the epitome of programming "cool." Therefore, Transmeta is cool.

"For a start-up, Transmeta has unusually high popularity," says Unix consultant Tim Berger. "This is partly due to Linus' popularity and his
freely available operating system. Linux rules. I think the general perception is that something different is brewing at Transmeta, something far more interesting than the average start-up," says Berger. "This perception is probably a reality."

Linux rules because to millions of geeks it represents an alternative to the Windows way of doing things, the Bill Gates stranglehold over the computing world. Transmeta is basking in Torvalds' reflected grandeur -- and not wholly without reason. Many Transmeta programmers are active on Linux-related mailing lists and newsgroups. Transmeta has even taken over the job of hosting the Web site where updates of the core source code of Linux -- the kernel -- are made available for download. Torvalds has also made it abundantly clear that he wouldn't have accepted the job at Transmeta if he hadn't been guaranteed the right to continue working on Linux.

Some Linux devotees hope, then, that somehow Transmeta might be able to do for hardware what Linux is doing for software. Together, Transmeta and Torvalds could free the world from the "Wintel duopoly" -- the domination of the entire computing industry by Microsoft and Intel.

No wonder then, that a shock went through the Linux community when the news spread across the Net on April 1 that Microsoft had bought Transmeta. Sure, most readers quickly realized the prank was an April Fool's joke, but the underlying anxiety still rang true. What was Torvalds doing in the cruel world of Silicon Valley, where dreams are bought and sold every day via stock swaps, and the goal of many a start-up is simply to become successful enough to be bought by Microsoft?

After all, Allen, co-founder of Microsoft, is a major investor in Transmeta. One worried Linux fan even theorized a "nightmare scenario" in which Transmeta is actually controlled by Allen, and Gates had asked his old high school buddy to make Torvalds "an offer he can't refuse."

Sounds crazy, huh? Torvalds seemed to think so.

"Where do people find these rumors?" he posted to Usenet. "Paul Allen is a stockholder, and he just happens to be one of the ones you find when you do a search for [Transmeta] on the Net. But 'controlled by'? Not even close."

Most likely there is no nefarious, Redmond-hatched plan to neutralize Torvalds' threat to Microsoftian world domination by locking him up in a hush-hush start-up thousands of miles from his native Finland.
There are certainly plenty of reasons why a company working on a dramatically new chip might covet Torvalds' expertise. As chip designer Wharton observes, it's increasingly impossible to separate the worlds of hardware and software. Plugging into the Linux community could give any new hardware product a leg up on its competitors.

At the very least, Transmeta has reaped massive publicity benefits simply by hiring Torvalds. Michael Learmonth, a reporter for the San Jose Metro who wrote a major feature on Torvalds a year ago, recalls that Transmeta executives seemed very "excited" at the prospect of a cover story on Torvalds. But after much deliberation, they would only allow the following information to be released about Linus' role at Transmeta: "Linus Torvalds works for Transmeta. Transmeta is a corporation located in Santa Clara."

"It seemed to me that they were trying to generate a little bit of heat just because they had a big name in Linus Torvalds," says Learmonth. "They knew that it was getting notice."

And notice carries market value -- regardless of what Transmeta's business turns out to be. Suppose Transmeta does announce a new product, and then immediately decides to go public and offer stock for sale. The scene is straight out of a wacky Silicon Valley novel: A horde of free-software fanatics rushes to buy the stock, making Transmeta employees and investors instant millions.

If the story of Transmeta proves anything, it is that fiction and reality are merging fast out there on Freedom Circle.

Or are they? We won't know for sure anytime soon. In an e-mail message, Transmeta CEO Ditzel predicted that Transmeta wouldn't announce a product for at least a year, and told me that "your readership might find our product plans a bit mundane." He noted that "we had a major change in direction a few months ago, and that has slowed us down a bit."

Slowed us down from what? Ditzel wouldn't say anything more.

Ditzel has a stellar reputation in the chip business. Formerly the director of Sun's Sparc Laboratories, two decades ago he was one of the earliest proponents of the RISC architecture for chip design -- a major breakthrough in computing. To this day, the Sparc workstation is one of the few alternatives to the Wintel duopoly -- leading some Transmeta observers to wonder if he's plotting another major chip breakthrough.

By their own admission, as evidenced by job postings to Usenet newsgroups, Transmeta is a "fabless semiconductor corporation" that plans one day to ship "a revolutionary new product." "Fabless" means that Transmeta doesn't have a fabricating plant -- it just designs chips. Allen's Wired World site adds another nibblet of information: Transmeta is supposedly working on "Alternative VLSI engines for multimedia PCs." This, however, like Transmeta's own name, is so vague a description as to be useless. VLSI ("very large scale integration") refers to chips with thousands of electronic components -- in other words, just about any chip on the market. And any new PC is a multimedia PC.

EE Times columnist Wolfe says that Transmeta has had several changes of direction: "Initially, industry scuttlebutt had them doing a PowerPC clone. Then it was a Java processor. Now, they may be planning a media chip which can tie into network-computing environments."

There is some hard evidence that Transmeta is working on chip designs for graphics processing. Transmeta is a member of VESA, the Video Electronics Standards Association, as well as AGP-IF, the Accelerated Graphics Port Implementers Forum. Then there's a reference at the Elpin Systems Web site suggesting that Transmeta "makes things that go faster." Elpin's president, Larry Coffey, refused comment, citing an NDA, but Elpin's main business is manufacturing test equipment for graphics hardware.

Both Brookwood of Dataquest and Wharton believe that Transmeta must be working on a chip that is in some way compatible with the "x86 architecture" that is at the heart of all IBM-compatible PCs. As Brookwood notes, this is a $23 billion-a-year market, 80 percent of which is controlled by Transmeta's neighbor, Intel.

"Just as Willie Sutton robbed banks because that was where the money was, x86 is where the money is, " says Brookwood.

The available evidence, then, suggests that Transmeta is working on some kind of superfast video processing chip that will plug into IBM-compatible PCs.

One senses Pynchon suddenly becoming bored. A chip that provides speedy video processing isn't exactly "revolutionary," nor is it the kind of breakthrough that will galvanize millions. Why the elaborate secrecy?

Of course, all this could be clever disinformation meant to distract avid Web surfers from Transmeta's secret plan to turn Silicon Valley on its head and topple Intel by pioneering a dramatically new chip architecture. Wharton himself, an Intel veteran, has dreamed of such a breakthrough. The key, says Wharton, would be devising some kind of superfast new chip that would allow for dramatic performance gains with new programs, but at the same time be able to run all your old Windows "x86" programs.

"The sort of thing I would do if I was in Dave's shoes would be to design a processor that can run existing x86 programs somehow, though perhaps suboptimally," says Wharton, "and then have a native mode where it runs the program with balls-to-the-walls optimum performance. These two designs aren't incompatible."

Such an approach might explain the hiring of Torvalds. Suppose you created a new ultrafast graphics chip, Wharton surmises, "and you were able to use the Linux community to port Unix operating systems and compilers to this new architecture. Instantly you would have access to the entire body of Unix software that is out there. If you are throwing a new chip out on the market, it would make immense strategic sense to involve the Linux community in your efforts."

Even more intriguingly, Wharton suggests that the approach he outlines would mean that the very name Transmeta would actually be "perfectly reasonable."

"What are the qualities of this approach to processor design?" asks Wharton. "You start with a meta architecture that's at a meta level compared to the programs that you are trying to run and then you take your existing programs and translate them into the existing meta architecture on the fly."

I asked Dave Ditzel what the name Transmeta signified.

"The name Transmeta has no particular meaning," answered Ditzel. "When you register a company, you need a name that isn't in use. It's surprisingly hard to find one. We just took prefixes and suffixes and combined them until we got a name that was unique."

Oh well. Never mind. Of course, if I were running in paranoid mode, wouldn't I expect Ditzel to say exactly that? We'll just have to wait and see.

But for how long? Transmeta was founded in late 1995. It received upward of $12 million in capital in 1997. Michael Learmonth, the San Jose Metro reporter, recalls that a Transmeta executive suggested they were on the verge of a major announcement -- almost a year ago. Three years is a long time for a start-up in Silicon Valley to go without bringing a product to market. As Dataquest analyst Brookwood notes, "The last company that was this secretive was MicroUnity [which promised a breakthrough "media processor chip"]. And they spent an enormous amount of money and produced nothing."

But Wolfe disagrees with the hypothesis that Transmeta's delays to market signify trouble.

"I think they're proceeding with the deliberate caution that befits a company intent on being a player for the long haul," says Wolfe. "Dave Ditzel won't let his company crash and burn the way MicroUnity did. Indeed, that's precisely the reason Transmeta's game plan has shifted so many times ... Ditzel's sort of like a surfer, trying to catch just the right technology wave and ride it for all its worth."

"If I was handicapping, I'd bet on Transmeta," says Wolfe. "I expect them to succeed because they realize that incremental progress -- taking one small step forward at a time -- pays off much more handsomely than the promise of 'revolution.'"

Except that, as least as proven by its job postings, Transmeta is promising revolution. And Linux geeks all over the world are praying for it. They'd do well to ask themselves the same question I asked in Transmeta's parking lot. Are those trees for real?


By Andrew Leonard

Andrew Leonard is a staff writer at Salon. On Twitter, @koxinga21.

MORE FROM Andrew Leonard


Related Topics ------------------------------------------

Linux Silicon Valley