Why it's not all plain sailing for Crusoe

Crusoe, the smart processor which could revolutionise computer design and help slay chip giant Intel, may not lead to the dawn of a new technological age.

If the chip fulfils all of its claims then it will be a major achievement. The hard part, however, will be getting partners to make it in production volumes to create products people are willing to pay for.

Turning technology on its head?
Transmeta, the company behind the design, claims that the software based chip, which can run any x86 compatible software, consumes about one quarter of the power of an equivalent Pentium. It is also cheap.

Crusoe is expected to appear in products by the middle of the year after more than four years of secret development. Some analysts are claiming that it will turn technology on its head.

The chip is also causing a lot of excitement because Linus Torvalds is part of the development team.

A chip off the old block
Analysts generally think that Crusoe is an impressive piece of technology. Clive Longbottom of Strategy Partners says that even if Crusoe can do half of what is claimed, it will be an incredibly important development.

"But if it turns out that the software is full of bugs then it could be the shortest lived 'great white hope' in history," warned Longbottom.

Part of the process
Instead of running the entire processor in hardware, the chip is surrounded by a software layer. The software is written in very long instruction word (VLIW) processing which packages a number of simple instructions and sends them to the chip for a single operation.

Crusoe's VLIW delivers four 32-bit operations in a single piece of information called a 'word'.

VLIW was pioneered by Multiflow and has been partly used in the next generation Intel chip, Itanium. But Crusoe's software has an additional function; it acts as a translator between the chip and the application. It optimises these translations for the Crusoe hardware and stores the results for re-use.

The upshot of using such software is that the Crusoe hardware does not need so many transistors, making the chip faster and crucially, meaning that it does not need much power to operate. This makes it ideal for portable use where battery life is the difference between a successful and an unsuccessful product.

Chips away!
Two Crusoe chips have been announced: the TM3120 for a 333Mhz chip for $500 to $900 portables and Internet devices, and a much faster TM5400 for a 700Mhz chip for $1,200 to $2,500 laptops.

It appears that the TM5400 will be configured to handle Windows, while the lower speed TM5400 chip will handle Linux. Torvalds has already stated that he considers Linux should be aimed for growth in the non-PC appliance market and that the TM5400 should fit that space nicely.

However, George Weiss, vice president of research at analyst GartnerGroup, points out that Torvalds seems to have very little control over the final product.

"It is a very proprietary product and all the rights are reserved by Transmeta. That state of play is a long way away from the great spirit of Linux open source philosophy," he said.

Transmeta will not even allow its customers access to the software code so that they can configure it for themselves. Crusoe comes pre-configured for Windows or Linux.

Stormy weather ahead
As the dust settles after the chip's launch, the industry is predicting stormy weather ahead for Transmeta. Some analysts claim that there is more than twice the possibility of compatibility and performance bugs in Crusoe than conventional chips.

Weiss said Crusoe could dent the sales of Intel's lucrative ARM chips, and certainly the chip giant's stock did dip in response to the Crusoe announcement. However, as Weiss pointed out, the market is still waiting for Intel's response. The chip giant could, for example, announce developments of its own which would leave Crusoe with stiff competition.

One industry source has indicated that Intel may even try to block Crusoe in the courts. Crusoe is shielded from the problems that blighted other attempts to emulate Intel chips because it is software rather than hardware based. Intel's patents only apply to hardware and Crusoe is designed differently. However, there is talk that Intel may try and block Transmeta from supporting the MMX standard SIMD.

Another problem is that there is a question over how much it will cost to get the product onto the market and whether or not original equipment manufacturers will support it. IBM has agreed to make Crusoe, but a similar deal with Cyrix was cited as one of the reason that company had difficulty competing. Device makers would have to adopt the chip and the process could take time.

Testing times
There is also strong competition in the handheld market with companies such as Hitachi, MIPs and Intel. Longbottom does not expect major computer manufacturers to support Crusoe until it had proved itself. He said the chip is still in the "smoke-and-mirrors" stage of marketing and that there has been no independent testing.

"We will probably be given a few 'demos' by the company for a couple of months, but we need to see the chip used in anger. In its raw state, it will probably be quite fast, but once it is in a device it could slow considerably," he said.

"Crusoe can be competitive if it can manage speeds between 600MHz or 700MHz, any lower and it will be competing with stuff which is already on the market."