Features

Efficeon no rival to ARM, yet, outside PC arena

Imagine you have a choice between two pocket computers. One is based on the ARM 11 core; the other is based on the x86 instruction set. OK, which do you pick?

The answer is all to do with power consumption.

At this week's Microprocessor Forum, the launch of a really low-power PC engine by Transmeta - the Efficeon processor - was put into perspective by another event at the same show - the launch of two new processors in the ARM 11 family.

Both chip families aim to reduce power consumption. The difference is simple: the Transmeta chip will run software for a PC - Windows, Linux and so on - while the ARM family will handle the software found in most mobile phones, pocket PCs and Palm machines.

If you have software written for a Windows notebook, then you'd have to re-write it before you could release it onto the ARM family; and you'd pay for it, heavily, in battery life.

"The ARM1156T2-S and ARM1156T2F-S cores will be the first CPUs released that integrate ARM Thumb-2 core technology extensions," said ARM at the show.

The release was instantly supported by Motorola, which announced that it was licensing the new designs "for the next generation of mobile phones and PDAs."

Thumb-2 is aimed at reducing memory footprint; you can build a smaller, less power-hungry machine with the same performance."Thumb-2 core technology is a blended 16-bit and 32-bit instruction set architecture for low power, high performance and small footprint code, which delivers software solutions that are 26 percent smaller than existing ARM solutions and 25 percent faster than existing Thumb solutions," said ARM.

That's uncannily similar to the message from Transmeta.

In its presentation at the MPF, it showed benchmarking slides comparing its new technology with Intel Pentium M processors (the chip inside Centrino computers). But it wasn't just comparing work done per MHz; rather, it looked at heating. The phrase "Thermal Design Power" was introduced: "What MHz can you achieve within a 7 Watt TDP?" asked Transmeta founder Dave Ditzel.

Different processors, he pointed out, achieve different MHz for the same TDP limit "To compare performance, you must compare at the maximum MHz for a given TDP limit" - and the seven watt limit is pretty much accepted as a reasonable one for a genuinely portable notebook. "After that, you need a fan."

It was a compelling argument. The processor inside the Efficeon is actually not a 32-bit processor. It isn't even a 64-bit processor; it's a Very Long Instruction Word processor of 256 bits; and it emulates an Intel x86 processor - very fast indeed. And if you find the "MHz to MHz" comparison meaningless in this context, then it had equally impressive performance tests.

Nobody expects Intel to watch while its Centrino brand is attacked. Centrino has impressively reduced the battery drain for PC notebooks, and the Banias research centre which produced it is certainly working equally hard on the next generation (expected next year) which should improve performance, and cut power consumption yet further.

But nobody is planning to build either a phone, or a PDA, using the Transmeta Efficeon, nor with the Centrino. Not even Intel is planning that.

The closest we ever came to seeing a cut-down PC in pocket form, was the Oqo design. Some raved about it when it was first announced in May 2002; this analyst thought it was a daft notion . "Nobody seems to be willing to rain on this parade, so I guess it's my job. The point I'm more concerned about is the fact that when Oqo runs on its built-in screen, it is 640 by 480 pixels. The phrase "power and richness" takes on an ironic tinge in that context," was my analysis.

It was worse than I thought. The company remains a startup, and remains pre-launch. The problem is simple: if you want a big, Windows level screen, then it consumes enough power on its own to make the power drain from the disk and the processor relatively small. But when you get down to pocket sized then it's perfectly feasible to have a half-pound weight machine that will run all day.

In the case of a smartphone like (for example) the Treo 600, it will run for a couple of weeks as a computer, and give you five solid hours of phone talk time, too.

No processor based on the x86 instruction set has yet been thought of as a rival to the ARM - not even by Intel. The only reason you'd use an Intel standard processor is to run Windows, or Linux.

In the next generation of Linux, with the expected explosion of Linux smartphones, there will probably be more devices running Linux in the pocket, never mind on the desktop, than running Windows.

When Microsoft wanted to take Windows to the pocket level, it re-wrote it. When Intel wanted to make a smartphone running Windows, it used that version of Windows - Windows Mobile - and it ran it on the XScale processor - an ARM derivative.

Does this mean that the days of the X86 processor are numbered? Hardly! - or if they are, it's far enough in the future that it would be difficult to pinpoint. There is always a demand for sheer, raw power - processing speed - and when you've got twin or quad P4 or Opteron processors running at 3 GHz or more, no ARM design is going to challenge it - especially at running Windows!

But what it does mean is that if you take Transmeta and Intel's claims to battery stamina seriously, it's got to be in the context of Windows. When it comes to a system that has to run all day and fit in the pocket, then ARM is in a different league.

Nobody is going to launch a pocket PC to take seriously, based on Windows and X86. It's a good joko, Oqo, but not something anybody will actually use.