Decoding Centrino

Intel stands on standards. What's the point of the P4-M now? Techical treats. Hotspot mania. And more...

Got Tips?

Analysis There's no doubt that Intel's Centrino platform is impressive technology. Exactly how much performance it delivers over existing processor solutions, and how much longer it will allow mobile users to operate on battery power we'll see in the coming months as Centrino notebooks are put through their paces.

And all the indications are that Intel will enhance the benefits Centrino brings in the second half of the year when Dothan, the successor to the 1.3-1.6GHz Pentium M chip that sits at the heart of Centrino ships. Dothan is being debugged right now, with early versions rolling off Intel's 0.09 micron/90 nanometre production lines.

But while Intel is revelling in the "revolution" that it believes the launch of Centrino will bring, and is rightly proud of the technological achievements it has wrought, in its bid to harness the wireless networking trend to pull the company and the PC business out of the doldrums, it's engaging in an almost Microsoftian attempt to dominate the market.

Usurping the standard

Centrino comprises an energy-efficient microprocessor, low-power support chipset and a plug-in 802.11b adaptor. IEEE 802.11b has been gaining interest since the late 1990s, not least since Apple introduced it to the mainstream in July 1999 with the introduction of AirPort, its branding of the standard.

Apple at least has always acknowledged AirPort's foundation on a standard. Not so Intel. Certainly, when asked, company representatives acknowledge the fact that it is, but at no point during Centrino's coming out party in London last night were 'Wi-Fi' or '802.11' mentioned.

That might be dismissed as the enthusiasm of the company at the launch of its latest product. But this unwillingness to mention the 'w' word extends to Intel's hotspot roll-out programme. Intel's vision is to have sites co-branded by Centrino, Intel and the network operator. The name of the actual standard on which this is all based doesn't get a look-in.

The concern is that the growing population of notebook users - and it is a growing market segment, one of the few (if not the only) expanding areas in the IT business - will associate wireless networking not with an open, interoperable standard but with a vendor-specific brand.

The thought won't be 'I need a Wi-Fi-compatible system to connect to the Internet on the move' but 'I need a Centrino system to connect to the Internet on the move'. That's the power of branding.

Hotspot? What spot?

That said, Intel does have some way to go. Despite showing a long list of Wi-Fi hotspot providers, coverage is a very long way from achieving the near-ubiquity of mobile phone coverage. A single 3G base-station will give the same coverage as 500 Wi-Fi base-stations.

Hotspot networks began appearing in the US about 18 months ago, as far-sighted airport authorities connected their lounges to the Internet via 802.11 base-stations. Last year, co-operative networks, such as Joltage and Sputnik, offered base-station owners roaming rights if they open their WLANs to other network members. At the same time, companies like Boingo, Gric, HereUAre and iPass began setting up base-stations and signing roaming agreements to allow members of other networks access to theirs.

Such activity seems to have been the sign that showed Intel there was more to Wi-Fi than new technology, that there was real user interest in connecting to the Internet almost anywhere. It began exploring hotspot partnerships a year ago, hopping onto the bandwagon early but by no means at the start. From the spin, you'd think Intel had invented all this stuff.

And following the British Government's decision to allow commercial services to operate in the unlicensed 2.4GHz band, similar services have begun to appear in the UK.

Intel's own figures for March 2003 show just 250 publicly available hotspots in the UK - one for every 220,000 members of the British population. Relatively few are mobile computer users, but it shows folk will have a long way to go to get public WLAN access today. For Europe as a whole, the hotspot total runs to a few thousand.

The wireless world Intel envisages is some way off yet, no matter how well Centrino sells, but the risk to the perception of the standards that make that world possible remains.

The P4-M question

Will Centrino sell well? Certainly it's an attractive proposition, on battery life and performance alone. Running flat out the 1.6GHz Pentium M can out-perform a 2.5GHz Pentium 4-M by around 12-15 per cent. Early independent reports suggest up to seven hours of battery life.

That begs the question: why buy a P4-M now? Intel's answer is that Centrino will appear in mobility-oriented notebooks, while the P4-M will still have a role to play in systems designed for portability - full-feature notebooks and desktop replacements. With the desktop P4 carving a share of the latter market, particularly at the lower end of the market, the space left for the P4-M is diminishing. Why choose a P4-M now?

A 1.6GHz Centrino will set you back around $1699; a 2.4GHz P4-M a little less than that, depending on specification. Intel says the P4-M will be around for some time, powering full-function notebooks, but if Centrino's better, what's to stop OEMs offering full-function Centrino notebooks? The power conservation advantage would be reduced, but buyers would still get the performance benefits.

We can envisage Intel bumping up the speed of the P4-M line to take it past Centrino's 12 per cent lead, but the P4-M's pricing compared to higher clocked desktop P4s, suggest that the latter and the Celeron will become the chips of choice for big, portable PCs, while Centrino will dominate mobility-oriented designs. Unless Intel slashes P4-M prices.

Confronting the Megahertz Myth

The other thorny question it faces is the Megahertz Myth. If a 1.6GHz Intel chip is faster than a 2.4GHz Intel chip, what does that say about the use of clock speed ratings as the definitive measurement of performance?

When we discussed this with Mooly Eden, head of Intel's Israel development centre and the guy in charge of the design of the Pentium M and the 855 chipset family that supports it, it was clear Intel wants to have its cake and eat it too.

Megahertz is not the only criteria for performance, he said, lauding the Centrino. Megahertz will always be crucial, he noted later, when questioned about the value of the P4-M post-Centrino.

In short, buyers who think megahertz is important will want the higher-clocked chip, and Intel's not going to stop them buying one. Smarter customers will choose the lower-clocked but faster chip. How many potential buyers will turn their nose up a Centrino, however, because they've had years of indoctrination? Centrino's other benefits may be enough.

Let's get technical

How have those benefits been achieved? In addition to a much more highly stepped (five to seven steps, says Eden) SpeedStep system, the Pentium M SpeedSteps the core, not the processor I/O, which is down to (and fixed at) 1.5V in any case. That, says Eden, provides the OS with much greater scope to drop frequency and voltage than the P4-M offers. Pentium M's well-publicised ability to put unused core components to sleep, and do the same with segments of the 1MB L2 on-die cache help too.

Interestingly of the 77 million transistors on the die, fewer than 27 million are dedicated to logic - the rest comprise all that cache.

Then there's the system bus, operating at 400MHz like the early P4, but pared down to allow a reduced operating voltage. Pared down how? By stripping out multi-processing capabilities, for a start. Even if there was room in a 'thin and light' notebook's shell, don't expect dual-processor Centrino boxes for some time yet.

Dothan is unlikely to change this, but the shift to 90nm fabrication will allow the Pentium M team to add all the features they left out of the first chip in the series. Eden insists Dothan is not just a die-shrink, though that surely will allow Intel to ramp up the clock speed without sacrificing battery life too. The 90nm fabrication process allows Intel to punch out 400 Dothans per 12in wafer. But don't ask how many of those actually work... ®


Biting the hand that feeds IT © 1998–2020