15. Transmeta
 | The material found below represented speculative research into just what Transmeta might be up to based on information findable from open sources (i.e. - on the Internet). With the product announcement in January 2000, this material is rendered largely obsolete, and it is fair to say that some of the web links are no longer active. It may still be interesting to see how close my speculations were... |
 | The purpose for creating this page, initially, was the rampant paranoia that surrounded the hiring of Linus Torvalds. Many wondered, at the time, just what this Transmeta company was all about, and some wild ideas came out. I did some research, the results of which are presented here, to determine what could be found from open sources. I've received phone calls from people trying to figure out more about Transmeta. Please don't bother. I don't have any inside information at all. This was an exercise in research. You would now do far better to contact Transmeta directly. |
I think the official word is still just "stuff". And if anybody asks "What kind of stuff?" the answer is "cool stuff". | ||
| --Linus Torvalds | ||
There was, at one time, slightly more information findable at Paul Allen's "high tech investment" web site. (Paul Allen was one of the founders of Microsoft, and is heavily into venture capital investment activities.)
Transmeta Corporation
CEO: David Ditzel
Primary business: Alternative VLSI
engines for multimedia PCs
Headquarters: Santa Clara, CA
Investment date: 1997 |
"David Ditzel is president and CEO of Transmeta Corporation, a fabless semiconductor company dedicated to high-performance VLSI solutions. He was previously at Sun Microsystems, where he was chief scientist for the SPARC Technology Business and director of SPARC Laboratories. "
See: TECHSTOCKS info " By unveiling the JavaEngine 1 at a high-profile industry event like the Java Internet Business Expo, Sun is sure to steal the spotlight for a few weeks' time. After the glow wears off, the company will face continued competition from a host of vendors aiming silicon at the network-computer market. Leading that list are IBM Microelectronics, with its PowerPC 403GX and Digital Equipment Corp. with its StrongARM processor. Waiting in the wings is startup Transmeta Inc. (Santa Clara, California.), which was co-founded by Sun alumnus and one-time SPARC architect Dave Ditzel. Transmeta is reportedly at work on a Java-specific microprocessor. "
I would, based on all of this, hazard the guess that Transmeta is in the process of designing a CPU designed for use in Network Computers.
The idea being to have something that provides decent performance with a minimum number of gates. (In several senses of "gates.") I would expect to see something with a somewhat SPARC-like architecture.
What with there being quite a lot of participants in this area, Transmeta would need to have some "in" to encourage third-parties to go with their product. Having a Linux port to the "TransParc chip" (I'm being a little facetious in the naming) from day one would be pretty beneficial, even if only as a platform for doing native development.
The usefulness of Linus in all of this should be pretty obvious. Having someone on hand who has been involved with writing an OS kernel that has been ported to a variety of architectures is really quite valuable.
The article The Red Herring - Stalking Transmeta - September 1998 found nothing of any value.
15.1. Other Visible Transmeta Activities
There have been some visible things coming out of Transmeta:
EDA on Linux initiative It's time to convince Electronic Design Automation (EDA) companies to get serious about supporting Linux. Performance of EDA tools on a Linux-powered PC is comparable to a proprietary Unix box, but at a fraction of the cost. Some small EDA companies already offer their products on Linux, yet the more established firms do not support Linux and few plan to do so. Why not? The big companies' standard reply is:
You can make a difference:
There's also a web site devoted to EDA on Linux | ||
| --Fergus Slorach on Wed, 23 Sep 1998 15:11:36 -0700, wrote: | ||
It sounds to me like people at Transmeta wouldn't mind seeing Linux EDA tools. Hmmm...
There have been quite a number of "Transmeta Speculations" that have amounted to either overjubilant or ravingly-paranoid fantasies, including (apparently a dead link) The Linu(s)x Apocaplypsis (A Transmeta Fantasy)
Rather strongly in support of the notion that Transmeta may assortedly be involved with building CPUs that might emulate the IA-32 architecture comes this patent filing: Memory controller for a microprocessor for detecting a failure of speculation on the physical nature of a component being addressed (US5832205)
Note that the patent discusses quite extensively methods of running software in an "emulation" mode, with the IA-32 architecture being the area of particular interest. The issue that causes emulation to be "hard" is that of memory-mapped I/O. An emulation scheme may run very quickly, if instructions may be reordered. I/O is an example of an application where this sort of reordering will make programs operate incorrectly. The patent is directed towards detecting these sorts of errors.
Presumably the approach would be:
Run the software in "reordering mode," full steam ahead.
When an I/O access is about to be handled incorrectly due to reordering, detect it before damage is done, and "sync up."
One of the most entertaining spurious rumours is that the New, Improved Amiga will use a CPU produced by Transmeta.
It is just plausible enough that I could believe it, and would not be overly shocked should the claim prove to be true. There is, however, no proof, and there is an ample availability of unsubstantiatable "hype" as to cause great entertainment and loud publicity to go in both directions.
Crusoe Exposed: Reverse Engineering the Transmeta TM5xxx Architecture I
Crusoe Exposed: Reverse Engineering the Transmeta TM5xxx Architecture II