Who can keep up with the milestones in scientific computing?

A decade ago, when I was a researcher in the field of computational physics, my graduate advisor had grants that allowed us access to fairly large supercomputers. The computers I used for my research were nowhere near as fast as the new petaflop-class supercomputer built at Indiana University. Ars Technica just recently published a quick look at this machine, which is the first petaflop computer at a university.

At that time, the big iron was at places called MSRCs – Major Shared Resource Centers. You had to pay for time on those machines – and for many of them, get security clearance, since they were owned by the military. (For example, my work was mostly done at the MSRC at the Naval Oceanographic Office.) By the time I graduated, universities were just starting to build their own in-house supercomputers.

Back then, supercomputers were used to perform fairly specialized types of calculations. I was doing molecular dynamics calculations – essentially solving Newton’s 2nd Law (F = ma) in its differential form for a quarter of a billion atoms simultaneously. Other folks were doing climate modeling – a different set of differential equations, or modeling processes for the Strategic Stockpile Stewardship initiative. (i.e., nuclear weapons.) These days, there are a host of bioinformatics applications that can use this horsepower and even more social science applications that all fall loosely under the heading of Big Data. With each increase in the number of floating point arithmetic operations per second (FLOPs), the size and complexity of problems that can be tackled also increases.

Despite Ars Technica’s flashy title for their piece, Indiana University certainly isn’t the first to build an in-house supercomputer, and certainly not the first to get plaudits for having a Top 25 supercomputer. My own alma mater was briefly the owner of a Top 10 supercomputer, in the 2002-2003 timeframe. They are just now installing the 3rd generation of that system, which clocks in at 146 teraflops, or about 10% of the IU machine’s speed.

While IU has the fastest supercomputer in a public university now, the rate of progress here will quickly eclipse their computer. In the relatively short span of 11 years, the fastest computers in the world went from about a teraflop to about a petaflop. That’s almost a hundredfold increase in speed per year. But the pace of progress doesn’t change the fact that with the increase in computing power at their disposal, researchers at IU now have the capability to solve these larger, more complex problems. And they can do it without having to get a grant of either money or computing resources. That’s a big deal for the progress of science, even if the superlative is fleeting.