Breaking the Petaflops Barrier

Last week, the US Department of Energy announced the first supercomputer to achieve a petaflop of sustained performance.  It will be housed at the Los Alamos National Laboratory in New Mexico.  The machine designed and built by IBM, is named Roadrunner, after New Mexico's state bird. 

A petaflop is a million billion calculations per second, that is, a 1 followed by fifteen zeros.  That is how many calculations per second Roadrunner can perform.  When talking about petaflops, the numbers are so large that it is hard to comprehend what they mean.  We are almost into numbers of astronomical dimensions. 

The IBM press release used a few analogies to describe the power of Roadrunner, such as "The combined computing power of 100,000 of today's fastest laptop computers";  and, "It would take the entire population of the earth, - about six billion - each of us working a handheld calculator at the rate of one second per calculation, more than 46 years to do what Roadrunner can do in one day."

The previous major milestone for supercomputers was the teraflop - which is a 1 followed by twelve zeros.  Crossing the teraflops barrier was a huge deal for the supercomputing community when we did it in the late 90s.  And here we are - only ten years later - with a machine which is 1000 times more powerful than a teraflop machine.

In my opinion, the most important explanation in the press release to help us appreciate Roadrunner's achievement is: "A complex physics calculation that will take Roadrunner one week to complete, would have taken the 1998 [teraflops] machine 20 years to finish – it would be half done today."  I may add that a calculation that would have taken the teraflop machine a week to complete, can now be done in 10 minutes in a petaflop supercomputer.  Doing something in a week instead of twenty years, or in 10 minutes instead of a week is a huge deal indeed.

How does Roadrunner do it?  Technology has advanced a lot in ten years, but clearly nowhere near 1000 times.  The innovative architecture of Roadrunner takes advantage of some very powerful market forces.  Roadrunner uses about 6500 dual core AMD processors, the kind used in high end personal computers.  But the extraordinary power of Roadrunner comes from its 12,240 Cell processors. 

The Cell processor was originally designed by IBM, Sony and Toshiba for the Playstation 3 game player.  It is noteworthy that IT is now so integrated into all aspects of society, business and our personal lives that the most advanced technologies are now coming from the world of consumer electronics - e.g., mobile devices, HDTVs and video game players.  The use of commercial technologies in Roadrunner will make it possible for IBM to sell supercomputers based on its architecture for all kinds of scientific and business applications.

The US Department of Energy has been funding the development of the world's fastest supercomputers as part of its National Nuclear Security Administration (NNSA)  initiative.  The NNSA is responsible for the US stockpile of nuclear weapons.  These weapons were mostly produced 30 - 40 years ago, and given the 1996 Comprehensive Nuclear Test- Ban Treaty, their reliability must be certified without conducting nuclear tests.  It takes a huge amount of supercomputing power to accurately simulate the extraordinary complexity of nuclear weapons systems, especially as you try to understand what happens in the first fraction of a second during an explosion.  This is why DOE needs advanced supercomputers like Roadrunner.

Beyond its military applications, leadership in supercomputing is viewed around the world as a symbol of national economic competitiveness and of technical and scientific leadership.  When a Japanese machine was ranked as the top performing supercomputer for several years starting in 2002, many articles were written bemoaning the loss of US technical prowess and prestige.  The articles only stopped when IBM's Blue Gene gained the number one spot in 2004, a position Blue Gene is now passing on to Roadrunner.

The advent of petaflop supercomputing will enable major advances in applications of all sorts, - in medicine, science, engineering and business.  Los Alamos researchers have already started to use Roadrunner in a project called Petavision that simulates extremely complex neurological processes.  Petavision models the human visual system, mimicking more than 1 billion visual neurons and trillions of synapses.  Applications like Petavision are critical to help us better understand the structure of the brain, which hopefully will lead to breakthroughs in treating major disabilities like Alzheimer, autism and schizophrenia.

Petaflop supercomputing will help us explore the major challenges associated with climate change, making it possible for scientists to test global climate models with far higher accuracy than has been possible so far.  We can expect advances in a wide range of application of critical importance to society, such as the development of biofuels, the design of more fuel efficient cars, personalized genomics-based medicine, and the ability to better understand and manage the behavior of our global, integrated – and increasingly unpredictable - digital economy.

For me personally, the Roadrunner announcement is a source of great pride.   Supercomputing has been a major part of my professional career, from my work toward a doctorate in physics - when I realized that I enjoyed the computing part of the research more than the physics and became a computer sciencist, - to the years I was general manager of IBM's fist parallel supercomputing initiative - the Scalable POWERParallel family - in the fist half of the 90s.

I am really proud of the achievement of my IBM colleagues who developed Roadrunner.  And, I feel really good about the innovation energy that Roadrunner is about to inject into the world's supercomputing community – urging them on to tackle grander challenges and achieve greater breakthroughs. 

Now, on to the exaflops machine, that is 1,000,000,000,000,000,000 operations per second.  We won't have to wait all that long . . . . .

Comments

[Chris Ward]

So far, there is no 'commercial software' for these kinds of machine. It's all either 'secret software' (as in the NNSA's application), or 'open-source software' scratched together by researchers on top of what universities have written over the decades.

The machines are there, and provision of the 21st-century equivalents of Microsoft Windows and Microsoft Office must be a business opportunity for someone.

And it's not just the IBM machines. I think Intel have the same problem ... or opportunity ... over what to do with their 80-processor chips from their research labs.

[Jordan Hydro]

Floods, droughts, pollution and climate change have created even more problems. It is important that China play an increasingly prominent role on climate change. right??