IBM's supercomputer chip breakthrough more PR than IT

[UPDATE x 2 - Yep, I got it embarrassingly wrong. The chief researcher on the project emailed me with the actual paper and some (very restrained, in the circs) explanation of what's actually going on. Look for a follow-up, just as soon as I've cleared up a couple of points I'm still unclear about (my understanding of optical physics is being, ahem, upgraded), but meanwhile please enjoy the post below for entertainment purposes only! Production, can you colour my rotund features a tad more rosy? Thanks...]



Oh, for heaven's sake.

Today, I got a press release from IBM - and, by the look of things so did everyone else with typing fingers. Here's the title and the first few lines:

New IBM Research Technology Could Enable Today’s Massive Supercomputers to be Tomorrow’s Tiny Computer Chips

Advancement in using light instead of wires for building supercomputers-on-a-chip

LONDON, UK, December 6, 2007 - Supercomputers that today consist of thousands of individual processor "brains" connected by miles of copper wires could one day fit into a laptop PC, thanks in part to a breakthrough by IBM scientists announced today.

And while today’s supercomputers can use the equivalent energy required to power hundreds of homes, these future tiny supercomputers-on-a-chip would expend the energy of a light bulb.

In a paper published in the journal Optics Express, the IBM researchers detailed a significant milestone in the quest to send information between multiple cores - or “brains” - on a chip using pulses of light through silicon instead of electrical signals on wires.

The breakthrough - known in the industry as a silicon Mach-Zehnder electro-optic modulator - performs the function of converting electrical signals into pulses of light. The IBM modulator is 100 to 1,000 times smaller in size compared to previously demonstrated modulators of its kind, paving the way for many such devices and eventually complete optical routing networks to be integrated onto a single chip. This could significantly reduce cost, energy and heat while increasing communications bandwidth between the cores more than a hundred times over wired chips. "

And, sure as eggs are ovoid avian reproduction devices, the headlines are already appearing across the Web. "IBM Researchers Build Supercomputer-on-a-Chip" is typical, but if you put "ibm supercomputer" into Google news, you'll find over four hundred similar statements.

It ain't so. It ain't even close.

What IBM has done is publish a report that its researchers have built a single optical component which uses silicon to modulate light at 10 gigabits a second. That's it. And, although the release presents this as a great breakthrough, Intel demonstrated something extremely similar - using exactly the same principle, and also on silicon - in July. Running four times faster than IBM claims. I know. I saw it at IDF in September. I talked to the researchers.

At the time, I did some checking to see if it really was the first, as they claimed, and at the time it seemed as if it was - although there were plenty of other researchers in America, Japan and Europe working in the field and things were progressing apace everywhere.

And if you talk to Intel about what's going to happen to this technology, they'll tell you that while it is extremely significant and has masses of potential, the practical problems of integrating it on-chip with mass-production CPUs are also extremely significant. It does not of itself usher in a bold new world of supercomputing.

Now, it could be that IBM's device really is '100 to 1000 times smaller' than the Intel device - I can't say, because there's no technical information in IBM's press release - but given the physics of how silicon Mach-Zenhder modulators work, I very much doubt it.

[UPDATE - the UK IBM PR has pointed me at a diagram of the key component of the IBM device. The Intel equivalent pic is buried in a PDF, page 15, so I can't link directly to it. The key dimension of the IBM device is about 400nm, compared to Intel's about 500nm, which I make around twenty percent difference - but they're basically the same thing and basically the same size.]

And all that stuff about reducing consumption from megawatts to laptop-watts? Sure, this will save some power, but it won't account for more than a fraction of the power savings of the overall systems, when they turn up.

I've asked IBM for clarification, in case I'm missing something embarrassingly obvious - and if I am, I'll be happy to 'fess up and get embarrassed.

But on the face of it, what we have is one extremely hyped-up press release and an entire Web of people happy to take it at its word rather than hit Google - or, heavens to Betsey - actually call a competitor to ask if this is actually true.

It's not as if you actually have to know what a Mach-Zenhder modulator is: the rule is, if someone claims to be first with something or have a 'breakthrough' technology - especially if every other word in the release is 'could' - you go and check. It's one of the oldest PR tricks in the book. You can see why.

The people who'll feel this the most keenly are the researchers, who'll have to go to conferences of their peers and explain that no, they didn't tell IBM's press office to print all that guff. That they've done good and important work, I do not doubt. But this will have done their reputation among their fellows no good at all.