Slashdot Mirror
Transmeta's Demise Predicted
egdull writes: "According to this story, Transmeta's party is over. Between buggy first-implementations of chips, leadership shake-ups, and "being outfoxed by Intel," Transmeta is done, according to C|Net. With a low stock price, they might be a target for a takeover, with Via being the only named interested party."
I wonder what Linus is thinking of doing if his employer goes.
A feeling of having made the same mistake before: Deja Foobar
Drives and Displays are.
So basically they came to market with a nice sounding product, but it was still a product that sells stock, not laptops. It was a product that used important keywords, claimed it could beat intel, and enlisted the god of Linux.
In the end its a product which really doesn't bolster laptop life all that much, and its real use was to make Intel provide the product that they could but did not have to.
You cannot taunt Intel or Microsoft, they have too many people with very large egos, and they will stomp you if you try. The best bet is to operate under their radar... and not to draw attention to yourself with brash claims versus these two behemoths until you can sustain your business.
* Winners compare their achievements to their goals, losers compare theirs to that of others.
Granted I am not an MBA nor a chip engineer, so this may be just wishful thinking, but I always wondered why Transmetta didn't play to the strength of their chip: i.e. you could make it act like other chips thru firm/software. I realize that x86 was where the market was, but I find it hard to believe that they wouldn't expect Intel to counter them in the marketplace (as they did).
I always thought that they should market it as an embedded chip, the lynch pin being they could supply you chips that wouldn't require you to relearn a new instruction set. I.e. if you're used to programming a Mips, they'd ship you the chip with the Mips instruction set. If you programmed PPC, then they'd ship you that. That would also give companies exposure to the underlying archetecture of the chip and maybe they'd migrate to its native instruction set.
Like I said, I'm but a mere code jockey, so what do I know.
Transmeta simply failed to deliver sufficient innovation to be competitive. The code morphing was an interesting idea, but they didn't do anything groundbreaking with it. Similarly, Intel managed to narrow the power consumption gap, while still beating them on the benchmarks.
They designed a chip for a market that doesn't exist - on the embedded side, processors like the StrongARM, SH3, and even, at the very low end, stuff like Z80's are smaller, cheaper, and lower power. At the same time, on the high end, ie. laptops, speed is king. With 15" LCD's on laptops these days sucking down the batteries, the power savings of the Transmeta chips weren't worth the lower performance, and certainly weren't going to help boost sales to mhz-obsessed consumers.
Ars has a good editorial in response to the CNET article.
At the end of my Crusoe article, I predicted that TM would eventually announce a workstation-class chip. It's been a little over a year and this still hasn't happened, but I remain convinced that they're working on just such a project.
Never the less, the transmeta processors are a good way to achieve binary compatibility using more advanced architectural concepts (VLIW). There lack of performance is a by product of the design decisions, if you want a low power synchronous design you have to reduce the speed [1]. The only suprise is that they achieved the performance they did vs. the difference in power consumption. Remember they are not just executing the bloated and inefficient [2] x86 instruction set but also effectively recompiling it as well. Personally I do not believe their ideas will dissappear just yet, though if the world economy goes into full recession Transmeta may not survive.
:-)
[1] I suppose you could increase the parallelism, but there is only so much you can get from the instruction scheme.
[2] If you do not believe it is inefficient then explain insructions such as XLAT (D7) and CMC (F5). (Yes i am sad enough to be able to assemble x86 code from memory
Physicists are said to stand on one another's shoulders while programmers stand on one another's toes.
Please keep in mind while reading the article that Intel was (and may still be) an investor in CNet. They may be hoping for a self-fulfillng prophecy with respect to Transmeta. Hopefully this is not the case and the article is fairly reported (I don't know enough about Transmeta to make that determination myself) - just be congnisant of the source.
No, Intel no longer owns any part of CNET. If they did, CNET would have included the statement indicating ownership.
I know this because a year ago or so, I noticed an Intel story that did not mention their relationship, and I contacted the author of the article. He responded that Intel had recently sold its shares in CNET.
"And like that
in addition to their virtualized x86 processor is one that supports many different architectures (Alpha, x86, PA-RISC, etc) on a single chip, with context switching between them. Add vmware to the mix, and you've got virtual OSs on top of virtual, native processors. Talk about being able to run platform 'A's native code on platform 'B'!
Tiller's Rule: Never use a word in written form that you've only heard and never read. You will end up looking foolish.
why does transmeta have to enter every market? they could easily license the technology to amd and remain a small profitable company. is the future really about wintel everywhere? look at arm who has a rather nice set of tech. large and unprofitable is really no fun!
The really sad part is that they stressed the "low power" part of their design. The really cool part was the code morphing. The crusoe was a full 128bit (256? can't remember, although they had it planned) processor that had an emulation layer rinning on it to translate x86 commands into it's own instruction set. This is really cool. Transmetta could change the whole chip design drastically and still maintain compatability with this layer. also you could speed up your computer with a simple bios update when they had finished more research for algorythms and tweaked the code some more.
I was wondering if any other company had interests in code morphing technology of this type.
WikiAfterDark.com It's a sex wiki, go now!
Embedded systems engineers are running into trouble for certain applications. On the one hand, StrongARM, MIPS, and even low-end microcontrollers like the 8051 are reliable and cool. But in the race for speed everyone has thrown power consumption to the wind. What if you need something more powerful than a StrongARM, but can't make use of Ultra SPARC and Intel processors because they simply run much too hot for embedded devices? So the embedded engineers are starting to resort to custom-designed processors tuned for specific purposes. But it would be much better if someone put effort into higher performance CPUs that didn't munch up 50 watts of power.
Maybe they ought to be translating Java byte codes. You're not going to beat raw x86 hardware with emulation, but you can beat other systems at emulating the Java VM.
Transmeta has about $262 million in cash, but it expects to burn through $20 million in the current quarter.
at that rate, if their business doesn't pick up, they'll be out of business in 6 years!
Unless you are a day trader, this shouldn't pose a problem. There is an element of risk involved, especially when investing in the short run. Looking in the long run, however, one can stand to make a great return on their investment. Transmeta isn't done; they have newer processors planned. Manufacturers have thus far only focused on the low-power properties of the Crusoe, and not very much has happened with the code morphing, which is evident from the C|Net article. Code morphing is a potential technology; when companies realize the value of it, then you will see manufacturers flocking over it rather than dropping it like a bad dream.
Considering that most of Transmeta's research over the past few years have been on concepts such as code morphing, and considering that Intel and AMD haven't been researching this venue, Transmeta would therefore have the lead in such a category for quite a while. The only question is: when will companies realize the value of this technology?
Note: I am not an analyst nor an investor (no money here). Feel free to take my argument with a grain of salt.
... then maybe I'll finally be able to put a decently fast (~300 mhz) CPU into an ATX case that doesn't need a CPU fan that keeps dying. Oh how I wish there was more low-maintinance hardware out there...
Examples:
Digital makes an Alpha chip that's 25% faster than Intel's chip. That's a noticeable speed boost! ... But, if you wait a year, Intel's chips will match its speed. So you might as well buy an Intel chip now and plan to upgrade in a few years.
Centaur makes a chip that's 25% cheaper than Intel's chip. That's a nice price drop! ... But, Intel makes so many chips that don't turn out to be 1.5 GHz P4s, it can afford to send out all those low-speed Celerons at roughly the same price as Centaur. So, you might as well buy a low-cost brand-name Intel chip.
Now, Transmeta makes a chip that's 25% cooler... and once again you can buy an Intel chip that's almost as good, but much more available.
In each of these cases, Intel has been able to shift the price-performance ratio and knock out a competitor. Only AMD's Athlon line, which is capable of competing with Intel from top to bottom, seems to be able to stake out its own territory.
I think the niche market for general purpose CPUs doesn't exist.