Sequoia A Step on the Path to Exascale Computing
Articles,  Blog

Sequoia A Step on the Path to Exascale Computing


>>Sequoia is the fastest computer in the world, installed at Lawrence Livermore National
Laboratory, its 96 racks of IBM Blue Gene/Q with an astonishing 1.6 million processor cores. Sequoia produces 16 petaflops of computation;
that’s 16 quadrillion operations per second. It’s an important stepping stone
on the way to exascale computing — machines that will be 50 times
as fast as today’s fastest. At Lawrence Livermore, Sequoia is used to safeguard the readiness of
the nation’s nuclear arsenal.>>Obviously we’re very, very pleased
that the Sequoia system is rated as the most powerful computer
in the world today. The last 10 years or so has been characterized
by a series of partnership investments with IBM in which we delivered not
just the Sequoia system but three generations of Blue Gene systems. The reason why we invested as we did
was that we recognized fairly early that the power requirements — the electrical
power grid, I’m talking about now — was such that our bills for paying for the use
of these machines was going to exponentiate and we wouldn’t be able…my entire
computer center budget would end up going to paying the electric bill, which
is hardly what any of us wants to do. So, when IBM first approached us with this idea
of the Blue Gene system, we rapidly engaged and we invested together and
essentially designed this system together, and what you see today is the outcome of that. Each of these systems has had a
quantitative…has represented a quantitative improvement over the one that
preceded it, and not a factor of two or three but often a factor of 10. And when you see a factor of 10
in computational improvement, what results from that is a qualitative
difference in the kind of science that you can do and the kind of
results that you can generate. So, it basically changed
the nature of our program.>>If I had to pick one thing that we
focused on when we were building Blue Gene/Q that made it truly unique, I would pick
on our bidirectional toroidal network. This is a way to connect the 100,000
processors of Sequoia together in a manner that was really codesigned with the National Lab
to match the way that they accessed the data. The kinds of communication that they needed
between data we built into the hardware of this machine to move that
data as efficiently as we could; to get it there as reliably as we could. And in the end we created a synergistic machine that has obtained the world’s highest
computational performance per watt that has ever been achieved.>>As we approach this notion of exascale
computing, it will open up a world of additional opportunities that to
this point really have been inaccessible to us by and large. And it’s not just science, per se; it’s an
investigation of a variety of applications across a wide-ranging set of
disciplines that will derive benefit. And of course we see it in areas like
materials science, where this notion of looking at the behavior of materials from atoms all
the way up to the physical implementation of the material give us insight into how
these materials might be employed in a variety of different kinds of applications
ranging from consumer products all the way up to the most modern aerospace
and automotive kinds of designs.>>Now, the problem with going to exascale
and beyond is basically one of physics. In some senses, everything we’ve done
up to date has been relatively easy. We’re reaching a kind of physics dominated
threshold in the design of microprocessors and in the design of systems which is
essentially going to stagnate progress. One of the problems is that if we just go off
and use the technologies that we use today, an exascale class computer would
require maybe 100 megawatts to run. That’s about $100 million a year. In the absence of concerted investments in integrated solutions,
progress is going to stagnate. And so it’s for this reason that many
laboratories and the DOE are working together with the Department of Energy and NSA
and in the Office of Science to see if there isn’t some national program that
could be initiated that will allow us to work with American industry in order for us to tackle
these problems systematically and perhaps reach at this…reach this exascale level some time in
the next decade or perhaps a little bit further.>>In the annals of computing,
we’ve come a mighty long way since Babbage’s Difference Engine Number 2. Sequoia is the latest amazing step
forward, but we still have a long way to go.

16 Comments

Leave a Reply

Your email address will not be published. Required fields are marked *