Transmeta ;)

[Al Hudson <nospam [at] york.ac.uk>]

Hi everyone..

Following various discussions we've had about Transmeta, both on-list and
in pubs in Sheffield ;), I thought I'd throw together my interpretation of
today's news for your delicatation...

Firstly, a little summary for those who haven't been following. Transmeta
today announced the release of two processors, the 3120 and 5400. The 3120
runs up to 400MHz, and is pretty much available now. That's what they were
demonstrating. The 5400 will come later, and will go up to 700Mhz.

These units are designed primarily for mobile use, although the technology
itself is not necessarily geared to mobile computers. The chip is
basically made up of a small (3-unit) arithmetic unit, and some execution
logic. However, 75% of this chip is made of software, and it is the
software which is the most important part of the chip.

The whole technology is based around compiling machine code from any
instruction set on the fly to the native 128-bit instruction set. This
compilation is then cached and reused. The software section is reinvoked
if more instructions need compiling, or if the processor thinks it can do
a better job.

There's a whole load of interesting stuff that this technology does. The
power management is extremely interesting: for example, where a standard
CPU reduces speed by 10%, they usually reduce power by 10% also. A
Transmeta processor will reduce power by 30% over the same speed
reduction. There's a load of interesting possiblities that the processor
opens up, as the processor is able to tell how much work is needed, and
dynamically vary the amount of work it does to compensate.

The technology itself, though, is arguably more interesting. The idea of
decoupling the hardware from the software seems extremely powerful.
Instruction sets can be switched between on the fly, so you can have many
programs running with different instruction sets. So while this means,
currently, the Transmeta chips are aimed at mobile x86 compatible
applications, the actual theory could be applied to anything; making chips
which do a lot of floating-point scientific work, or graphics work, or
biasing the chips towards other things. It is possible to reduce code
complexity and size with customised intruction sets, there's a lot of work
being done in that kind of area... it's not just limited to mobile x86
clones.

To bring this all a little more on-topic, I have to say wow. Only a couple
of days ago I was talking about putting ELKSibo on my Psion 3a, and then
Transmeta go and show off 'Mobile Linux', a custom distro which will
enable people to use standard linux software on web-pads and the like.
Transmeta seem to be shipping a 'whole solution': the hardware, the
code-morphing software, the operating system, and in addition to this they
have a bevvy of reference hardware. I would be surprised if Crusoe
technology wasn't in the shops in a year's time. And it'll probably be
palm pilot like. I believe the Pilot has started losing ground to Wince
recently, perhaps it's just in time ;)...

We also might start getting stuff like WINE on non-x86 platforms..
personally, I can't wait to get my mitts on one and have a go. Not that
they'd be any different to anything else in normal use, but just to have
one ;))

Cheers,

Alex.

Start your own FREE mailing list at

© 2000 Microsoft Corporation. All Rights Reserved