Linux Today: Linux News On Internet Time.

VNU Net: Transmeta's Crusoe chip: survivor or castaway?

Mar 23, 2000, 19:43 (3 Talkback[s])
(Other stories by Neil Fawcett)

By Neil Fawcett, VNU Net

Every day a new technology comes along and attempts to change the way we handle data. Usually they die out, leaving the de facto standards such as Microsoft's Windows operating systems and Intel's x86 chip architecture to flourish.

In the world of mobile computing, there is a new pretender to the throne - a new technology, a new supplier, a new chip and a new concept to play around with. It's called Crusoe, and comes from a company called Transmeta. Don't feel too bad if you've never heard of it - it's early days for Transmeta.

What's so special about Crusoe?
Marketing people around the world will always ask the same things when it comes to new technologies: What's the unique selling point? Why should someone be interested in buying it?

The Crusoe chip, launched in January, is what its creator calls a 'smart processor'. Transmeta claims that this chip, the first in a series of smart microprocessors, will create a new category of mobile internet computers. It is supposedly based around a breakthrough software approach that will "revolutionise the field of mobile computing".

This, Transmeta claims, delivers on the market's need for 'all-day computing' with a PC-compatible solution that is unmatched in performance with low power. But for something to define a new category of anything you need companies manufacturing hardware around it, and you need a software industry to support it (see below for more on this).

The core design of the Crusoe chip is a superb piece of engineering work, and melds hardware and software into a pretty seductive package.

What you have is a radical departure from traditional microprocessor design in that Transmeta makes use of software to implement many functions you would normally implement in a hardware instruction set. It is this that gives Crusoe the claimed high performance and low power which is needed for demanding mobile computers.

At the heart of the Crusoe architecture is its code morphing software. Code morphing software surrounds a simple very long instruction word silicon engine to act like a fully compatible PC processing engine - in this case the software compatibility is with Intel's x86 instruction set. This works by using the software to 'morph' the x86 instructions into what the underlying hardware engine understands.

Transmeta describes Crusoe as a smart processor which 'learns' about an application while it runs, and then uses that experience to extend battery life. Using a proprietary power management technology called LongRun, Crusoe can be made to continuously adjust its operating speed and voltage to match the needs and workload of the application.

The technology can apparently make these adjustments hundreds of times per second, which Transmeta claims dramatically extends battery life. This contrasts with other processors which run at a fixed operating speed on batteries, wasting battery life on unused capacity.

There is also a multimedia aspect to LongRun which can be used to meet the demanding needs of rich data - such as video streams - which typically drain a PC's battery in as little as an hour. With LongRun, Transmeta claims it is possible to design a lightweight mobile PC which plays a DVD film for three hours or more.

Who's supporting Crusoe?
We asked Michael Dell, chief executive of PC manufacturer Dell Computer, for his view on new and emerging technologies such as Crusoe. A chief executive is seldom likely to lambast a technology out of hand, just in case his head of research and development suddenly shows up with a prototype, but Dell did have his doubts.

"We've been talking to Transmeta for some time now and it looks like great technology. But isn't it just emulation? And we all know the problem with emulation," said Dell. "There's been emulation on all platforms; some worked, some didn't."

Dell's point is valid. Transmeta can call this anything it wants to, but what it boils down to is that the morphing technology employed looks very like emulation - not something that turns the computer industry on.

Having heard Dell's views, we sought a couple more opinions - from Compaq and Hewlett Packard (HP). While neither company would dismiss Crusoe as a non-starter, both preferred to talk about Windows CE based hardware and Intel's efforts in the mobile arena.

HP is determined to bring CE-powered palmtop hardware, such as its own Jornada, into the enterprise as a device that can be managed and 'fed', meaning that data from the enterprise can be passed to such a device in a secure way. It sees the mobile device as a good corporate citizen, and not just a geeky toy that business people use to carry around telephone numbers.

This raises an interesting issue, and one that potentially benefits Crusoe. The demand from industrial and corporate users suggests that the software and data architecture of a mobile device, as in what it can do within an enterprise, is far more important than the hardware around which it is based.

If you extrapolate from this, then a chip from Crusoe, Intel, Hitachi, AMD or whoever, will suffice as long as it supplies what mobile computing needs - power without battery drain. There's no need for Intel inside.

In a mobile internet world, therefore, a Crusoe based computer has every chance to do well and carve out a niche. If you technically evaluate Crusoe, you find a chip that is much smaller than, say, a mobile Pentium II, needs less power and is x86 compatible via the Transmeta code morphing software layer that exists within the chip. Indeed, a Crusoe chip consumes one watt of power, which is about a tenth of Intel's best offering.

So is Transmeta's Crusoe really a threat to Intel? The best answer to this is no, based on the information available. If you look at the company's website - www.transmeta.com - you will see that few people have signed up as technology partners.

A big advantage that Crusoe has over Intel's design methodology is that potential bugs in the chips can be fixed faster and new features added more quickly. This can be carried out via software rather than hardware.

Instead of requiring a manufacturing change, as Intel's chips do, engineers can just write new software, test it, and then distribute the new code to a partner, perhaps over the internet. Or so the theory goes.

But back to Crusoe's future. Apparently IBM will be manufacturing the chip, which is good news because IBM's semiconductor fabrication people are highly experienced.

We will undoubtedly see different machines arrive based around the Crusoe chip, and these will fall into two camps: the low powered x86 Windows model and the low powered Linux mobile.

The first chip introduced by IBM is the TM3120, which is designed to act as the powerhouse behind mobile internet appliances, what people refer to as web pads, smart phones, and so on. It will use the mobile version of Linux as its operating system and will run at 400Mhz, with 108Kb of cache Ram, consume one watt of power and be x86 compatible.

The high-end counterpart to this chip is the TM5400, which is expected to run Microsoft Windows and will operate at a speed of between 500 and 700Mhz/s. It has 400Kb of cache Ram, uses one watt of power and is also x86 compatible. It is designed for lightweight notebook computers and anything which needs extended battery life.

But because of the Transmeta chip's design, it will not always operate at its highest clock speed. For example, let's say that a 700Mhz TM5400 is being used to play a DVD film - quite a common activity on notebook PCs these days. It does not take 700Mhz to play a movie, maybe as little as 500 MHz in fact, so the Crusoe chip is intelligent enough to scale itself back and therefore consume less power.

As regards pricing, Transmeta will cost between $119 (£75) and $329 (£208) for the TM5400, and $65 (£41) and $89 (£56) for the TM3120. How this will affect the street price for mobile units is anyone's guess.

Crusoe offers an insight into how mobile internet devices could look and operate. But it's a long way from market dominance yet.

Related Stories: