IBM Corp yesterday shook up the staid image of supercomputing, unveiling a new architecture for a next generation computer which it says has enough potential power to find a cure for diseases such as cancer and AIDS. The new computer, nicknamed Blue Gene is capable of processing one quadrillion operations per second, or one petaflop, which IBM says is 500 times faster than its predecessor, Deep Blue, and 40 times faster than the aggregate sum of all the world’s supercomputers today.
Unlike today’s models, which are used for a variety of purposes from predicting weather systems to powering and simulating the manufacture of car and aircraft designs, Blue Gene will be targeted squarely at modeling the complex nature of human biological protein design, which IBM sees as a huge potential revenue spinner and an as yet untapped market.
Speaking during a press conference yesterday, Paul Horn, senior vice president of IBM Research, said there are currently vast amounts of data being created by the various gene-mapping techniques, but that information technology was required to understand it. IT is becoming the language of biology, he said, but we can’t understand what’s coming out of these techniques without tremendous amounts of IT.
IBM’s plan is to model the manner in which human proteins build themselves into 3-D shapes and specifically to work out what it is that goes wrong in proteins that cause diseases, such as AIDs or cancer. Until now, there’s been no computer powerful enough to model the complex folding nature of the amino acids and IBM is positioning Blue Gene as the first computer to do that. IBM knows if its Blue Gene can come up with the answers, drug companies will pay billions to develop products to tackle them.
We’re going to unlock the secrets of how our bodies work, said Horn. We need to understand how they [proteins] fold and always get into the same shape. Frankly, we don’t have a clue today why or how that happens. He added: Our aim is…to understand why it folds in the wrong way sometimes to get, for example, sickle cell anemia. We’ll take those viral proteins and understand how they attack ourselves.
The project is costing IBM $100m in people and research to build what Horn describes as a fundamentally new architecture. Horn said that left up to Moore’s Law, it would take 15 years to produce a computer with the processing power of Blue Gene. But we want to achieve that in a third of the time.
The architecture it has come up with to do that is called SMASH (Simple, Many and Self-Healing). Using SMASH, Horn says that IBM has managed to drastically reduce the number of instruction sets per chip (from hundreds to just 57), combine memory onto each processor, reduce each chip’s power consumption and increase the overall yield of chips in the manufacturing process.
The result is a computer that will house one million processors and cost IBM $2,000 (and six gallons of water a second for cooling purposes) to run each day, according to Dr Monty Denneau, one of IBM’s key researchers behind the Blue Gene project. But despite having worked on the architecture for the last two to three years, the supercomputer is still another four to five years away.
Add that to the fact that the run time of a standard-sized protein modeling session will take one year running 24 hours a day, 7 days a week, and it’s obvious IBM isn’t going to come up with the cure for cancer overnight, but the financial implications are clear. Horn said that IBM is already partnering with many of the big names in pharmaceuticals, such as Aventis (the new technology arm of Hoechst) and Bristol Meyers Squib, as well as bio technology giants such as Monsanto and added that many of them are likely to make investments as the Blue Gene project gathers steam.
The combined revenues of the three key sectors that this kind of research is aimed at – agro-sciences, pharmaceuticals and bio technology – is around a third of a trillion dollars, said IBM’s Daniel McCurdy, vice president for Life Sciences at IBM, and while the IT budgets for R&D for those organizations is just $3bn, it would be fair to say that we see this as an opportunity bigger than just supplying the IT to these companies.
McCurdy agreed that joint ventures with pharmaceutical firms to model the interaction between viruses such as HIV, sickle cell anemia and human cells, could be business models that IBM would pursue, as well as the simpler idea of creating intellectual property and licensing it to drug firms. The two approaches are not mutually exclusive, he said.
In the meantime, IBM will use existing technology to begin work folding sets of smaller proteins, with just 30 or 40 amino acids. The aim is to produce a set of models as a reference bank for when the proper work of modeling the viral proteins gets under way.
Ten or fifteen years down the line, once Blue Gene is off the ground, IBM says the technology could potentially be put to other uses, such as looking at how galaxies are formed or even simulating the big bang itself to try and find out how the world was created. More down to earth though, it also reckons that the project will give Big Blue and its partners tremendous opportunities for IT which it hopes it will be able to scale down to solving everyday problems.
