Yorkshire Building Society (YBS)

2
Case study | Universit Paul Sabatier
Named for the 1912 Nobel prize-winning
chemist, Paul Sabatier, Universit Paul
Sabatier Toulouse III (UPS) has its roots in
one of the oldest universities of Europe—
Universit de Toulouse, founded in 1229. While
steeped in ancient tradition, the modern UPS
is world-renowned for its diverse array of
programs in the sciences, health, technology,
athletics, and engineering. It is also one of
France’s preeminent research institutions,
engaged in publicly funded advanced research
largely performed in partnership with major
scientic organizations.
One such organization is the Centre National
de la Recherche Scientique (CNRS), or national
center for scientic research. CNRS is the
largest governmental research organization
in France and the largest fundamental science
agency in Europe. Working in collaboration
with CNRS, the university conducts advanced
laboratory research in quantum chemistry.
Deep in UPS labs, researchers test new
theories in hopes of discovering scientic
insights that could ultimately be applied in the
real world, such as life-saving developments in
the pharmaceutical industry.
As part of its quantum chemistry research,
UPS uses a non-traditional method for
computing the electronic structure of
molecules with quantum Monte Carlo (QMC)
simulations, using “random walks” to solve the
Schrödinger equation (a dierential equation
applied to very large dimensional spaces)
rather than a traditional analytical approach.
While not new, the random walks method was
rarely used until recent years when massively
parallel high-performance computing (HPC)
systems made it more feasible. Now, more and
more researchers are using the random walks
method because it produces more accurate
results than traditional methods.
Each QMC simulation involves billions of
Monte Carlo steps, and can require thousands
of computing nodes to handle such CPU-
intensive oating-point calculations. Standard
x86 servers not only take up a lot of physical
space, they also consume enormous amounts
of power. Therefore, researchers at UPS turned
to a much more energy-eicient computing
architecture: HP Moonshot System.
Anthony Scemama, a research engineer and
HPC scientist with CNRS, is working under the
direction of Dr. Michel Caarel who is leading
the university’s quantum Monte Carlo project
now running on HP Moonshot. “We wanted
to nd a way to run these computationally
intensive calculations using as little energy
as possible,” Scemama explains. “So we
engaged a team of computer scientists to run
benchmarks to optimize the green eiciency of
our software code running on HP Moonshot. “
Supercomputing in a
fraction of the space
UPS deployed HP Moonshot with 45 HP
ProLiant m350 servers running Ubuntu Linux
in the HP Moonshot 1500 Chassis. With
eight cores per server, the ProLiant m350
is the densest CPU core count available in a
Moonshot System, enabling UPS to run a very
large number of low-frequency CPUs in a
fraction of the space compared to traditional
servers.
“We need to achieve as many gigaFLOPS as
possible while using very little power,” says
Scemama. “With HP Moonshot we get twice
as many gigaFLOPS per watt as traditional
servers. And with Moonshot’s dense scale-
out architecture we have no problem running
hundreds of nodes. It’s like having a micro-
supercomputer where we can develop and test