When Crusoe met SpeedStep, which left the bigger footprint in the sand?
There is a lot of excitement in the mobile chip arena at the moment. In the blue corner, Intel finally wheeled out its SpeedStep technology. The launch press conference didn't really throw up any major surprises other than how heavily Intel was pushing the mobile market as a whole.
The following day was more of a surprise. The latest adventure of Mr Linux, Linus Torvalds, is a fabless chip company, Transmeta. Torvalds' involvement ensured a startling degree of interest in the launch of its very first products, the Crusoe processor family.
On paper at least, these chips look interesting. First, there's the TM3120, in production now (it's being made by IBM) and aimed squarely at the Net device market. If these prove popular, then Transmeta could be in a very interesting position indeed, especially as the company also showed a Mobile Linux build aimed specifically at this market.
More importantly for the traditional mobile market, Transmeta's other chip, the TM5400, is aimed head-on at existing high-end mobile processors. With hastily-drafted quips mocking the Intel SpeedStep announcement, there's no doubt about just where Transmeta wants Crusoe to compete.
At the moment, Intel is the clear leader in the mobile space. Its chips are faster than the mobile K6-2 and K6-III - AMD provides the only real competition in notebooks - and its SpeedStep announcement was meant to consolidate this lead.
After the repeated delays and considerable flagging of the benefits that the product would bring, the announcement itself was a bit of an anti-climax. Fundamentally, a SpeedStep CPU is a low-voltage mobile Pentium III, available as either a 600 or 650MHz chip. The SpeedStep technology allows it to drop clock speed to 500MHz when running on battery power, saving power proportionately to the decrease in clock speed.
The clever bit is that core voltage is automatically reduced at the same time as the clock speed falls, from a peak of 1.9V to 1.35V. While this might not seem a lot, power consumption varies as the square of voltage, so such a reduction has a much larger effect. Intel claims that a 20% reduction in performance can be offset against a 50% power saving - a clear advantage if you need to use your PC on the move.
But in truth, this was quite a disappointing announcement. Given Intel's history of introducing new technologies first on a mobile platform (both 0.25 and 0.18 micron manufacturing were first seen in notebook chips, as were the Dixon-class chips with 256Kb of full-speed, on-die cache), a faster PIII that can slow itself down isn't exactly gripping.
No matter how much fuss Intel chooses to make about SpeedStep, the timing of the Transmeta announcement invites direct comparisons. In truth, Crusoe appears to offer much more of interest. Doug Laird, Product Development VP, concluded his presentation claiming 'we really did re-invent the microprocessor,' and while that reeks of marketing hype it's worth examining more closely.
In order to retain x86 compatibility, Crusoe is a combined silicon and software solution - Transmeta is determined to get across the message that it is a processor implemented as much in code as on a chip. In principle, its Code Morphing technology acts as a virtual machine, translating the existing code to Crusoe native instructions on the fly.
Windows and its applications should run with no need for recompilation - Transmeta rightly accepts that compatibility with existing software will be critical in driving acceptance of the new chip. What's more, it claims that this dynamic interpretation of code holds few performance disadvantages - it expects its 700MHz FM5400 chip to be comparable to an Intel PIII - and in fact can lead to more efficient execution, as code is optimised on the fly.
Performance and compatibility issues aside (Transmeta did demo Quake, and the TM5400 won't be available until the summer in any case), it's the LongRun technology embedded in Crusoe that appears to make it particularly suitable for mobile use.
Broadly, this equates to Intel's SpeedStep techniques of throttling back clock speed and simultaneously reducing core voltage. However, LongRun appears more sophisticated in the way it achieves its power-saving goals. Rather than being linked to the notebook's power source, LongRun seeks to reduce clock speed according to the needs of the active application. DVD playback, for instance (Transmeta sees constant processing load applications as a growth area for mobiles) requires a clock speed of around 400MHz - dynamically reducing speed to this from Crusoe's peak of 700MHz would offer substantial power savings in itself.
When this is combined with a core voltage that can fall to as little as 1.2V (Intel's minimum is 1.35V) Transmeta claims to be able to run a Crusoe chip at around a quarter of its peak power consumption. This has heat benefits too - a Crusoe operating temperature under load of some 48C is rather healthier than Transmeta's claim that a PIII under similar conditions will top 100C.
Unfortunately, measuring the real benefits this reduced consumption will offer to battery life is something of an inexact science. Accordingly, Transmeta has fallen back on the old standby of generating its own benchmarks that seek to combine power consumption and performance. Predictably, these results show Crusoe in a favourable light as against a (non-SpeedStep) PIII-500, with the new chip claimed to be between two and six times as efficient depending on the tasks performed. A pinch of salt may be needed - but such claims won't last beyond the first chip availability if they are overly exaggerated.
In fact, Transmeta has been incredibly bullish overall. CEO David Ditsel went so far as to claim "If it has a battery and a full-screen web browser, then it will be built with Crusoe" - aiming at nothing less than dominance of the mobile computing marketplace.
Announcing a 700MHz chip when Intel's mobile flagship offers only 650MHz is nothing less than a declaration of war. What's more, it carries the struggle into the top-range space that Intel has always seen as its own. Winning - or at least claiming to win - the battle of MHz is good for headlines, as AMD found with its desktop Athlon chips.
But this is really the first time that Intel has been challenged at the top end of the portable market. Its current competition primarily consists of AMD mobile processors, which combine lower speeds (under 500MHz) with relatively high power consumption. Their low cost has tended to win AMD market share in budget notebooks, at the end where design features are constrained by money and reducing heat output and power consumption is less of an issue.
At least until we see a mobile version of the Athlon, high-end notebook processors consist of Intel's PIII range, overshadowed though it may be by the looming silhouette of Crusoe. But though it talks fighting talk, Transmeta gave little indication of how exactly it intends to gain market share.
It must compete against Intel's brand name and image for acceptance by manufacturers and - probably even more important - OEMs. It's sure to be competitive on price in the same way as AMD, but Transmeta will have to deliver a pretty compelling price-performance metric to get market share. If it doesn't live up to most of its promises it will have an even more difficult task before it.
The company expects to carry all before it, but we fear Crusoe may end up like Betamax - leaner, meaner, arguably better - and a long long way from widespread adoption. Shades of Linux, anyone?
However, there is one particularly bright ray of hope. Building a new product brand and growing market share is always easier in an expanding market, and mobile computing is nothing if not a growth area.
Ironically, it's perhaps Intel's SpeedStep press conference that offers the most recent support for this view. The venue may have been dreary, but the presentation was notable for the lack of emphasis placed on the new SpeedStep chips. Instead, Intel concentrated on its expectations for growth in the mobile market. In essence, it told us how many chips it hopes to sell - and then pointed out that if they were SpeedStep-enabled they would be very profitable.
Unsurprisingly, notebook and peripheral manufacturers are keen to endorse this prediction of further exponential growth, although their perspectives vary in emphasis. Toshiba expects the market to keep growing at at least 20% per annum, and is comfortable that its sales will maintain their present momentum. At the moment, the bulk of Toshiba's business is with corporate customers, and the company expects this to continue. It doesn't - at least in public - anticipate a significant switch to aggressive competition at the lower end of the market.
Perhaps this view is coloured by its current relatively poor showing in retail sales. Toshiba would like to concentrate on the quality and the "overall value" offered by its notebooks rather than on the price tag, and wishes that buyers would too. It's a message that is easier to get across in a corporate environment than to a price-sensitive buyer in PC World.
By contrast, other manufacturers are more sanguine about the prospect of switching away from corporate sales into other areas. Xircom sees the biggest growth of its accessories coming from consumer or education buyers, while HP is keen to talk about the commoditisation of notebook PCs. While its 17% overall growth rate for next year isn't far from Toshiba's guesstimate, HP prefers to see the bulk of this growth being driven by low-end sales.
Historically, its base has been in corporate sales, where HP, as with others, has sought the prestige of large plum contracts. But with the recognition that these can be expensive to win - dedicated sales teams, heavy discounts on big orders - has come a fragmentation of the market.
Although each sale in a consumer environment or to the smallest SME customers may be worth very much less, a more distributed channel has lower sales costs. Equally, smaller sales carry smaller discounts and the potential for more solid margins.
Just as interesting for the customer is the ever-falling price of notebooks. Inevitably, it's a chicken and egg situation, but as technology has advanced, with faster processors and bigger screens, so notebooks have become more popular. Increased sales have led to a spiral of falling prices, so notebooks are now becoming more popular in education and in the home, especially where they may be being bought as a second or third PC.
Technology drivers, especially as they offer more performance and improved battery life, continue to make a notebook a more realistic proposition at the lower end of the market. A £1,200 machine, perhaps all that a consumer can afford, now competes on reasonably equal terms with a desktop PC.
What's more, it offers good quality to go along with its low price. According to HP, corporate clients are now picking up low-end notebooks in volume because they offer value at least as good as the more corporate products. Capital costs are much lower - and if a notebook is cheap enough to be treated as a commodity it can be painlessly disposed of when its life is over.
If it carries through, such commoditisation can only drive the notebook market strongly forwards. And an increased sales volume will drive the acceptance of new technologies - those same new technologies that have already broadened the market to its current level.