Intel’s top exec tells how zettaflop processing will revolutionize current computing methods by offering speeds one billion times faster than a current-generation systems.
Intel’s CEO Paul Otellini took the stage at Dell World to talk about what he sees as “the insatiable need for computing.” To put it another way, there are a number of tasks that Otellini – and Intel, as a whole – see as being mostly computationally restricted.
What that means is that they are limited by how much computer power we can throw at them, by how fast those computers are, and by how fast their chips can process and decode information. He brings up a number of different research activities that the company sees as changing within the next decade and a half, all thanks to the (so far) unrelenting forward process of Moore’s Law.
Moore’s Law, invoked first by Intel’s co-founder, Gordon Moore, states that the number of transistors that can be affordably placed within a specific area or chip roughly doubles every couple of years. While there is of course some overlap on what that means for the end user (end user in this case meaning anyone using the hardware to accomplish a task, from scientific researchers to a suburban soccer mom), these greater and greater numbers of transistors mean that chips are able to accomplish more and more tasks.
“Today,” Otellini said, “it takes a teraflop machine a few hours to diagnose a single cancer event from a high resolution CT scan.” A petaflop machine, the next generation, will take “seconds.” That wildly cuts down on how long a patient and their doctor have to wait in order to determine a future course of treatment.
On a closely related idea, more powerful machines will also bring down how much it costs to sequence personal genomes. When you are able to rapidly sequence your own complete genome, it opens a whole world in terms of personalized medicine.
Perhaps most interesting was Otellini’s idea of what a zettaflop machine will be capable of performing. One billion times faster than a current-generation teraflop machine, zettaflop processing could revolutionize how we model the weather.
One of the examples in the presentation talked about predicting the path taken by individual hurricanes and the spot where they would make landfall. For intense storm systems, that means preparation and often evacuation of the affected areas. Today’s computers can predict when the storm will hit land within a two-week period, across a distance of some 100 miles.
A zettaflop-capable machine, however, can cut that uncertainty by significant amounts. Instead of knowing when it will make landfill within a two-week period, we will know it within a two-day period. That one-hundred mile range of seaboard? Try being able to limit your prediction to a single zip code.
Lots and lots…and lots…of cores
Of course, predicting the future is heady stuff, and customers, journalists and scientists alike can all get wrapped up in the predictions and be very, very wrong. How does Intel plan on pushing the capabilities of these chips ever further? Well, that’s where Knights Corner, the new refresh of their MIC, or “Many Integrated Core” initiative, comes in. Due out next year, that platform will have chips with a whopping fifty cores on a single die.
While those chips will not be available for Joe Consumer on the street to go out and buy for his eXtReMe gaming desktop, corporations and research institutions will be buying machines from companies such as Dell that use these chips. Those machines will then be employed in some of the aforementioned tasks, such as gene decoding and weather modeling.
One of Dell’s first wins in that area comes in the form of helping to build the University of Texas’ new “Stampede System.” The SS will be based on Dell’s Zeus Servers, which use Intel Xeon E5 processors as well as some of the Knights Corner MIC microarchitecture – the whole system is expected to be capable of 10 petaflop performance. The system is expected to be completed in 2013.
Perhaps most exciting, however, is Intel’s prediction of where computers are going beyond all of these examples. It is said that we as a society are still at the beginning of what computers can do; looking at some of the charts and statistics, it is difficult to disagree. Estimates put the number of transistors around the world at less than 200 quintillion (one quintillion = one million trillion). By 2015, that number is expected to skyrocket to almost 1,200 quintillion.
The curve won’t stop there. Computers will become increasingly more powerful as time marches on, and while they are indispensable for many, if not most of the world today, it is difficult to imagine their ubiquity or their capability more than two decades from now.
Personally, I’ll settle for a computer that displays information along a pair of contact lenses (or even interfaces directly with the optic nerve). Detecting early cancers, though, well…that’s a good start.