Cerebras Wafer Scale Engine Outperforms World's #1 Supercomputer, Achieving Long-Timescale Molecular Dynamics Simulations 179x Faster
May 15 2024 - 9:00AM
Business Wire
Breakthrough Unlocks Millisecond-Scale Simulations for the First
Time, Enabling Scientists to Observe Never-Before-Seen Material
Behavior
Cerebras Systems, the pioneer in accelerating generative AI, in
collaboration with researchers from Sandia, Lawrence Livermore, and
Los Alamos National Laboratories, have achieved an extraordinary
breakthrough in molecular dynamics (MD) simulations. Using the
second generation Cerebras Wafer Scale Engine (WSE-2), researchers
were able to perform atomic scale simulations at the millisecond
scale – 179x faster than what is possible on the world's leading
supercomputer 'Frontier,' which is built with 39,000 GPUs.
Existing supercomputers have been limited to simulating
materials at the atomic scale at a rate of 2-3 simulated
microseconds per month, restricting our understanding of how
materials evolve and behave over longer periods. This breakthrough
achieved by Cerebras and its collaborators at national laboratories
has shattered this barrier. By harnessing the power of the Cerebras
WSE-2, the processor at the heart of the Cerebras CS-2 system,
researchers can now simulate materials for milliseconds – an
astounding leap that opens up entirely new vistas in materials
science.
“The partnership between the NNSA laboratories and Cerebras
Systems is part of the Advanced Memory Technology (AMT) program,
which aims to accelerate exascale supercomputers by 40x as early as
2025. With Cerebras’ currently deployed wafer-scale computers, the
teams achieved this materials science breakthrough and a speedup
that exceeded the goal of the AMT program by more than 4X,” said
James H. Laros III, Distinguished Member of Technical Staff at
Sandia National Laboratories and AMT program lead. “This experience
bodes well for future impacts to our program and potential
scientific advances.”
The research team innovated across both hardware and software to
overcome the limitations of today’s supercomputers. By mapping
individual atoms onto the WSE-2's nearly one million cores and
enabling efficient communication between neighboring cores, the
system simulated 270,000 timesteps per second across 800,000 atoms
– a staggering 179-fold speedup over Frontier, the world's leading
supercomputer. This breakthrough allows researchers to gain
unprecedented insights into the long-term behavior and future
evolution of materials at the atomic scale.
"This work changes the landscape of what is possible with
molecular dynamics simulations," said Michael James, Chief
Architect of Advanced Technologies and co-founder of Cerebras
Systems. "Simulations that would have taken a year on a traditional
supercomputer can now be completed in just two days. Scientists
will now be able to explore previously inaccessible phenomena
across a wide range of domains.”
Long timescale simulations will allow scientists to explore
previously inaccessible phenomena across a wide range of
domains:
- Materials scientists can now study the long-term behavior of
complex materials, such as the evolution of grain boundaries in
metals, leading to the development of stronger, more resilient
materials.
- Pharmaceutical researchers can simulate protein folding and
drug-target interactions over physiologically relevant timescales,
accelerating the discovery of life-saving therapies.
- Renewable energy experts can optimize catalytic reactions and
design more efficient energy storage systems by simulating
atomic-scale processes over extended durations.
To achieve this remarkable feat, the Cerebras team employed a
novel mapping scheme that assigns each atom to a single core on the
WSE-2. The cores are organized in a 2D grid, with the physical
simulation domain mapped onto this grid to preserve locality. Cores
communicate with their neighbors to exchange atom information,
allowing for efficient parallel processing. The WSE-2's unique
architecture, with its high memory bandwidth and low-latency
communication fabric, enables this fine-grained parallelism to be
exploited effectively.
“The NNSA’s Advanced Memory Technologies program started 1.5
years ago with the goal of 40x performance improvement on critical
NNSA applications over what can be achieved on exascale systems. We
all had our doubts on achieving this goal within the short
timeframe, but Cerebras’s technology and team has helped us exceed
this goal by demonstrating unprecedented 179x performance
improvement on MD simulations,” Siva Rajamanickam, Sandia National
Laboratories, Principal Member of Technical Staff. “These results
open up new opportunities for materials research and science
discoveries beyond what we envisioned. We are excited to continue
this collaboration with Cerebras and explore new frontiers in
science.”
As Cerebras continues to push the boundaries of high-performance
computing with its wafer-scale technology, even more groundbreaking
advancements are on the horizon.
“I have been working in atomistic simulation of materials for
more than 20 years. During that time, I have participated in
massive improvements in both the size and accuracy of the
simulations. However, despite all this, we have been unable to
increase the actual simulation rate. The wall-clock time required
to run simulations has barely budged in the last 15 years,” Aidan
Thompson, Sandia National Laboratories, Distinguished Member of
Technical Staff. “With the Cerebras Wafer-Scale Engine, we can all
of a sudden drive at hypersonic speeds. This joint DOE-Cerebras
team has accomplished something unprecedented and highly
disruptive. I look forward to seeing how this transforms all kinds
of scientific research in the near future.”
For more information, please see
https://arxiv.org/abs/2405.07898.
About Cerebras Systems
Cerebras Systems is a team of pioneering computer architects,
computer scientists, deep learning researchers, and engineers of
all types. We have come together to accelerate generative AI by
building from the ground up a new class of AI supercomputer. Our
flagship product, the CS-3 system, is powered by the world’s
largest and fastest AI processor, our Wafer-Scale Engine-3. CS-3s
are quickly and easily clustered together to make the largest AI
supercomputers in the world, and make placing models on the
supercomputers dead simple by avoiding the complexity of
distributed computing. Leading corporations, research institutions,
and governments use Cerebras solutions for the development of
pathbreaking proprietary models, and to train open-source models
with millions of downloads. Cerebras solutions are available
through the Cerebras Cloud and on premise. For further information,
visit https://www.cerebras.net.
View source
version on businesswire.com: https://www.businesswire.com/news/home/20240515983777/en/
Press contact (for media only) ZM Communications Email:
pr@zmcommunications.com