HPC 2022
High Performance Computing
State of the Art, Emerging Disruptive Innovations and
Future Scenarios
An
International Advanced Workshop
July
4 – 8, 2022, Cetraro, Italy
|
Programme
Committee |
||||||
|
L.
GRANDINETTI (Chair) University
of Calabria P.
MESSINA (Co-Chair) Argonne National Laboratory G.
ALOISIO University of Salento K. AMUNTS Juelich
Supercomputing Centre F. BAETKE European Open File System Organization P.
BECKMAN Argonne
National Lab. R. BISWAS NASA Ames Research Center C.
CATLETT University of Illinois System J.
DONGARRA University
of Tennessee S. S.
DOSANJH Lawrence
Berkeley National Lab. I. FOSTER Argonne National Laboratory G. FOX Indiana
University W.
GENTZSCH The UberCloud H. KOBAYASHI Tohoku
University T.
LIPPERT Juelich Supercomputing Centre S.
MATSUOKA Tokyo
Institute of Technology M.
PARASHAR Rutgers University V.
PASCUCCI University of Utah and Pacific Northwest National
Lab. T.
STERLING Indiana University R. STEVENS Argonne
National Laboratory W. TANG Princeton University M. TROYER Microsoft
Research |
ITALY U.S.A. ITALY GERMANY GERMANY U.S.A. U.S.A. U.S.A. U.S.A. U.S.A. U.S.A. U.S.A. GERMANY JAPAN GERMANY JAPAN U.S.A. U.S.A. U.S.A. U.S.A. U.S.A. |
|||||
|
Organizing
Committee L. GRANDINETTI (Chair) ITALY |
||||||
|
M. ALBAALI (OMAN) |
J. DONGARRA (U.S.A.) |
W. GENTZSCH (GERMANY) |
P. BECKMAN (U.S.A.) |
P. MESSINA (U.S.A.) |
R. STEVENS (U.S.A.) |
|
Sponsors
|
AMAZON WEB SERVICES |
|
|
CEREBRAS |
|
|
COLDQUANTA |
|
|
CSC Finnish Supercomputing Center |
|
|
CSCS Swiss National Supercomputing Centre |
|
|
DELL |
|
|
E4 Computer Engineering |
|
|
EOFS |
|
|
Hewlett Packard Enterprise |
|
|
IBM |
|
|
INTEL |
|
|
Juelich Supercomputing Center, Germany |
|
|
LENOVO |
|
|
NEXT SILICON |
|
|
NORTHWESTERN UNIVERSITY |
|
|
NVIDIA |
|
|
PARTEC |
|
|
PSIQUANTUM |
|
|
SAMBANOVA SYSTEMS |
|
|
SAMSUNG |
|
|
University of Calabria Department of Computer Engineering, Electronics, and Systems |
|
Media Partners
|
|
|
THE News Analysis Site for HPC Insiders: We’re all
about HPC, but like the emerging HPC ecosystem, insideHPC is evolving with a
number of important updates and topic focus areas coming later in 2022. To better serve our growing readership and
advertising base, insideHPC will deliver an updated format and featured
spotlight coverage of the emerging markets of enterprise HPC, HPC-AI,
exascale (and post-exascale) supercomputing, quantum computing, cloud HPC,
edge computing and High Performance Data Analytics. Our mission is to cover every aspect of
HPC. Written and edited by seasoned HPC journalists,
insideHPC represents the many voices of the global HPC community: where HPC
is today and where it’s going. We want to hear from you! Send story ideas and
announcements to news@insidehpc.com. We also encourage you to contact us with suggestions,
comments, corrections or ideas for stories, by emailing us at editorial@insideHPC.com. Or you can
contact our editor-in-chief, Doug Black, directly at doug@insideHPC.com Please follow insidehpc.com on LinkedIn, Twitter and
Facebook! |
|
UberCloud
provides Cloud With One Click – a fully automated, secure, on-demand,
browser-based and self-service Engineering Simulation Platform for engineers
and scientists to build, compute, and analyze their
engineering simulations. Our unique HPC software containers facilitate
software packaging and portability, simplify access and use of any public, privale, hybrid, and multi-cloud resources, and ease
software maintenance and support for end-users, IT teams, and their cloud
service providers. Please follow UberCloud on LinkedIn and contact us for performing
a Proof of Concept in the Cloud. |
|
Speakers JAMES AHRENS Los Alamos National Laboratory Information Science and Technology Institute Los Alamos, NM U.S.A. ILKAY
ALTINTAS San
Diego Supercomputer Center and Workflows for Data Science (WorDS)
Center of Excellence and WIFIRE Lab University of California at San Diego, CA U.S.A. FRANK BAETKE EOFS European Open File System Organization GERMANY PETE BECKMAN US DOE Argonne National Laboratory and University of Chicago and Northwestern University / Argonne National Lab. Institute for Science and
Engineering U.S.A. KEREN BERGMAN Electrical Engineering Department and Lightwave Research Laboratory Columbia University, New York U.S.A. BRENDAN BOUFFLER HPC Engineering Amazon Web Services U.K./U.S. CHARLIE CATLETT Discovery Partners Institute University of Illinois System U.S.A. CARLO CAVAZZONI SVP of Cloud Computing & Director of the HPC Lab Chief Technology & Innovation Office Leonardo S.p.A. Genoa ITALY ALOK CHOUDHARY Electrical and Computer Engineering Department Northwestern University U.S.A. TRISH DAMKROGER Hewlett Packard Enterprise U.S.A. EWA DEELMAN University of Southern California and Information Sciences Institute Los Angeles, CA U.S.A. JACK DONGARRA Electrical Engineering and Computer Science
Department and Innovative Computing Laboratory University of Tennessee Knoxville, TN, U.S.A. and Oak Ridge National Laboratory, U.S.A. and University of Manchester, U.K. DANIELE
DRAGONI Leonardo
S.p.A. Genoa ITALY ANDREW FELDMAN Founder and CEO of Cerebras
Systems Sunnyvale, California USA IAN FOSTER US DOE Argonne National Laboratory Data Science and Learning Division and Department of Computer Science at the University of
Chicago U.S.A. GEOFFREY FOX Department of Computer Science School of Engineering and Applied Science University of Virginia, Charlottesville and Digital Science Center Indiana University, Bloomington U.S.A. WOLFGANG GENTZSCH The UberCloud Regensburg GERMANY and Sunnyvale, CA USA VLADIMIR GETOV Distributed and Intelligent Systems Research Group School of Computer Science and Engineering University of Westminster London UNITED KINGDOM VICTORIA GOLIBER D-Wave Systems Inc. GERMANY and U.S.A. MARTIN HILGEMAN DELL Technologies U.S.A. BLAKE JOHNSON IBM Quantum Quantum Platform Lead U.S.A. CARL KESSELMAN Information Sciences Institute and Department of Industrial and Systems Engineering and Viterbi School of Engineering and School of Medicine University of Southern California Los Angeles, CA U.S.A. HIROAKI KOBAYASHI Architecture Laboratory Department of Computer and Mathematical Sciences Graduate School of information Sciences Tohoku University JAPAN KIMMO KOSKI CSC - Finnish IT Center
for Science Espoo FINLAND SALVATORE
MANDRA Quantum
Artificial Intelligence Lab (QuAIL) KBR, Inc. NASA, Ames Research Center Moffet Field, CA U.S.A. STEFANO MARKIDIS KTH Royal Institute of Technology Computer Science Department / Computational Science
and Technology Division Stockholm SWEDEN SATOSHI MATSUOKA Director RIKEN Center for
Computational Science, Kobe and Tokyo Institute of Technology, Tokyo JAPAN ELENA
MESSINA Principal, Prospicience LLC. U.S.A. PAUL MESSINA US DOE Argonne National Laboratory, Argonne Illinois Argonne Associate and Distinguished Fellow U.S.A. MASOUD
MOHSENI GOOGLE
Quantum Artificial Intelligence Laboratory,
Google AI GOOGLE Headquarters Venice CA, Los Angeles U.S.A. CHRISTOPHER MONROE IonQ Inc. and University of Maryland and Duke University U.S.A. MARTIN MUELLER SambaNova Systems Palo
Alto, California U.S.A. KEVIN OBENLAND Quantum Information and Integrated Nanosystems Lincoln Laboratory Massachusetts Institute of Technology MIT Boston, MA USA MANISH PARASHAR Scientific Computing and Imaging Institute and School of Computing University of Utah, Salt Lake City U.S.A. VALERIO PASCUCCI Center for Extreme Data Management, Analysis and Visualization and Scientific Computing and Imaging Institute School of Computing University of Utah, Salt Lake City and US DOE Pacific Northwest National Laboratory U.S.A. KRISTEN PUDENZ Atom Computing Berkeley, California U.S.A. DANIEL REED Department of Electrical and Computer Engineering School of Computing University of Utah Salt Lake City, Utah U.S.A. CHAD RIGETTI Founder and CEO Rigetti Computing Berkeley, CA U.S.A. MARK SAFFMAN COLDQUANTA Quantum Information and University of Wisconsin-Madison U.S.A. CLAUDIO SCALESE EuroHPC JU: HPC Research & Innovation LUXEMBOURG THOMAS SCHULTHESS CSCS Swiss National Supercomputing Centre Lugano and ETH Zurich SWITZERLAND JAMES C. SEXTON IBM Fellow IBM T.J. Watson Research Center,
N.Y. Future Computing Systems U.S.A. PETE SHADBOLT Chief Scientific Officer & Co-Founder PsiQuantum Corp. Palo
Alto, California U.S.A. GILAD SHAINER NVIDIA Mellanox networking at NVIDIA, focusing on high-performance computing, artificial intelligence and the InfiniBand
technology U.S.A. THOMAS STERLING AI Computing Systems Laboratory (AICSL) School of Informatics, Computing, and Engineering Indiana University, Bloomington U.S.A. RICK STEVENS US DOE Argonne National Laboratory Computing, Environment, Life Sciences Laboratory and University of Chicago U.S.A. FRED STREITZ Center for Forecasting and Outbreak Analytics (CFA/CDC) and Lawrence Livermore National Laboratory (LLNL/DOE) U.S.A. WILLIAM TANG Princeton University Dept. of Astrophysical
Sciences, Princeton Plasma Physics Laboratory and Center for Statistics and Machine Learning (CSML) and Princeton Institute for Computational Science &
Engineering (PICSciE) Princeton University U.S.A. MICHELA TAUFER The University of Tennessee Electrical Engineering and Computer Science Dept. Knoxville, TN U.S.A. SCOTT TEASE Lenovo Vice President HPC and AI Morrisville, NC U.S.A. PHILIPPE THIERRY INTEL Corporation U.S.A. ROBERT WISNIEWSKI Senior Vice President and Chief Architect of HPC Head of Samsung’s SAIT Systems Architecture Lab U.S.A. |
Workshop
Agenda
Monday,
July 4th
|
Session |
Time |
Speaker/Activity |
|
State of the art
and future scenarios |
||
|
9:30 – 9:45 |
Welcome
Address |
|
|
|
9:45 – 10:15 |
J.
DONGARRA |
|
|
10:15 – 10:45 |
T.
STERLING Transition to Memory-Centric
HPC Architecture for Data Analytics |
|
|
10:45 – 11:15 |
J.
SEXTON Accelerating Discovery |
|
|
11:15 – 11:45 |
COFFEE BREAK |
|
|
11:45 – 12:15 |
S.
MATSUOKA |
|
12:15 – 12:45 |
R.
WISNIEWSKI |
|
|
|
12:45 – 13:00 |
CONCLUDING REMARKS |
|
|
Emerging Computer
Systems and Solutions |
|
|
|
17:00 – 17:25 |
C.
SCALESE Presentation of the European High-Performance
Computing Joint Undertaking & its activities |
|
|
17:25 – 17:50 |
E.
DEELMAN 16 Years Later: Pegasus in Cetraro: challenges and solutions for emerging computing
systems |
|
|
17:50 – 18:15 |
B.
BOUFFLER We meet again at last. We’ve been busy |
|
18:15 – 18:45 |
S.
TEASE HPC and Sustainability: The Smarter
Path to Zero Emission Computing |
|
|
|
18:45 – 19:15 |
COFFEE BREAK |
|
|
19:15 – 19:35 |
P.
MESSINA Laudatio for prof. Jack Dongarra |
|
|
19:35 – 20:00 |
M.
HILGEMAN Innovating and democratizing HPC,
challenges for a platform vendor |
|
|
20:00 – 20:10 |
CONCLUDING REMARKS |
Tuesday,
July 5th
|
Session |
Time |
Speaker/Activity |
|
AI on HPC Platforms
I |
||
|
|
9:25 – 9:50 |
I.
FOSTER Intelligent Simulations: How
Combining AI and HPC Can Enable New Discoveries |
|
|
9:50 – 10:15 |
G.
FOX |
|
|
10:15 – 10:40 |
R.
STEVENS |
|
|
10:40 – 11:05 |
K.
BERGMAN Multi-terabit Photonic
Connectivity for Energy Efficient AI Computing |
|
|
11:05 – 11:35 |
COFFEE BREAK |
|
|
11:35 – 12:00 |
A.
FELDMAN |
|
|
12:00 – 12:25 |
P.
BECKMAN |
|
|
12:25 – 12:50 |
A.
CHOUDHARY |
|
|
12:50 – 13:00 |
CONCLUDING REMARKS |
|
|
AI on HPC Platforms
II |
|
|
|
17:00 – 17:25 |
c.
kesselman How not to Forget What You Have Learned: A
Data-Centric Approach To Reproducibility and Sharing for Machine Learning |
|
|
17:25 – 17:50 |
W.
TANG HPC Acceleration of Progress in
Fusion Energy Prediction & Control Enabled by AI / Deep Learning |
|
|
17:50 – 18:15 |
E.
MESSINA |
|
|
18:15 – 18:45 |
COFFEE BREAK |
|
18:45 – 19:45 |
PANEL DISCUSSION “Whither AI + HPC + Big Data?” Paul
Messina (chair) Preliminary
slate of panelists: Keren
Bergman, Andrew Feldman, Geoffrey Fox, Carl Kesselman,
Martin Mueller, Rick Stevens The use of AI has exploded in recent years. Many of the applications
that have been tackled make use of HPC architectures and algorithms, as well
as massive data collections. There has been much progress and experience
gained, which makes this a good time to assess what worked well and what did
not. The panelists will be asked to identify
advances in algorithms, software environments, data analytics, standards, and
computer architectures that would lead to more productive and reliable use of
AI. |
Wednesday,
July 6th
|
Session |
Time |
Speaker/Activity |
|
The QUANTUM COMPUTING
Promises I |
||
|
|
9:00 – 9:25 |
M.
MOHSENI Nonlocal Monte Carlo with algorithmic thermal and
quantum fluctuations |
|
|
9:25 – 9:50 |
C. RIGETTI |
|
|
9:50 – 10:15 |
C.
MONROE |
|
|
10:15 – 10:40 |
M.
SAFFMAN |
|
|
10:40 – 11:05 |
B.
JOHNSON Enabling quantum-classical
computation on multiple time scales |
|
11:05 – 11:30 |
COFFEE
BREAK |
|
|
|
11:30 – 11:55 |
K.
PUDENZ |
|
|
11:55 – 12:20 |
V.
GOLIBER |
|
|
12:20 – 12:45 |
K.
OBENLAND Benchmarking and Analysis of
Noisy Intermediate-Scale Trapped Ion Quantum Computing Architectures |
|
12:45 – 13:00 |
CONCLUDING REMARKS |
|
|
|
The QUANTUM
COMPUTING Promises II |
|
|
|
17:00 – 17:25 |
P.
SHADBOLT |
|
|
17:25 – 17:50 |
S.
MANDRA Large-Scale Simulations of Quantum Circuits on HPC
Clusters Using HybridQ |
|
|
17:50 –18:15 |
D.
Dragoni Exploration of QUANTUM COMPUTING @ Leonardo Labs |
|
18:15 – 18:45 |
COFFEE
BREAK |
|
|
18:45 – 20:00 |
PANEL DISCUSSION “The Intersection of Quantum Computing and HPC” Daniel Reed (co-chair), Christopher Monroe
(co-chair, t.b.c.) Panelists: Pete Beckman, Rick Stevens, Chad Rigetti,
Pete Shadbolt During the past several decades, supercomputing speeds have gone from
Gigaflops to Teraflops, to Petaflops and Exaflops.
As the end of Moore’s law approaches, the HPC community is increasingly
interested in disruptive technologies that could help continue these dramatic
improvements in capability. This interactive panel will identify key
technical hurdles in advancing quantum computing to the point it becomes
useful to the HPC community. Some questions to be considered: · When will quantum computing become part of the HPC
infrastructure? · What are the key technical challenges (hardware and
software)? · What HPC applications might be accelerated through
quantum computing? |
|
|
Is the “belle époque” of classical High Performance
Computer Systems coming at the end? |
||
Thursday,
July 7th
Friday,
July 8th
|
Session |
Time |
Speaker/Activity |
|
Key Projects, Novel
Developments and Challenging Applications |
||
|
|
9:30 – 9:55 |
K.
KOSKI |
|
|
9:55 – 10:20 |
K.
PUDENZ Long coherence times pave the path to quantum
applications |
|
|
10:20 – 10:45 |
V.
GETOV Component-Based Message Passing
for Java: Developments, Achievements, and Impact |
|
10:45 – 11:15 |
COFFEE BREAK |
|
|
|
11:15 – 11:40 |
W.
GENTZSCH |
|
|
11:40 – 12:05 |
T.
SCHULTHESS |
|
|
12:05 – 12:30 |
S.
MARKIDIS What we Learned About
Programming Models During the Race to Exascale |
|
12:30 – 12:40 |
CONCLUDING REMARKS |
Chairpersons
PAUL MESSINA
US DOE Argonne National Laboratory, Argonne
Illinois
Argonne Associate and Distinguished Fellow
U.S.A.
WOLFGANG GENTZSCH
The UberCloud
Regensburg
GERMANY
and
Sunnyvale, CA
USA
THOMAS STERLING
AI Computing Systems Laboratory (AICSL)
School of Informatics, Computing, and
Engineering
Indiana University, Bloomington
U.S.A.
EWA DEELMAN
University of Southern California
and
Information Sciences Institute
Los Angeles, CA
U.S.A.
GEOFFREY FOX
Department of Computer Science
School of Engineering and Applied Science
University of Virginia, Charlottesville
and
Digital Science Center
Indiana University, Bloomington
U.S.A.
HIROAKI KOBAYASHI
Architecture Laboratory
Department of Computer and Mathematical
Sciences
Graduate School of information Sciences
Tohoku University
JAPAN
IAN FOSTER
US DOE Argonne National Laboratory
Data Science and Learning Division
and
Department of Computer Science at the
University of Chicago
U.S.A.
VLADIMIR GETOV
Distributed and Intelligent Systems Research Group
School of Computer Science and Engineering
University of Westminster
London
UNITED KINGDOM
WOLFGANG GENTZSCH
The UberCloud
Regensburg
GERMANY
and
Sunnyvale, CA
USA
|
“Whither AI + HPC + Big Data?” Tuesday, July 5th, 2022 Panelists: Keren
Bergman, Andrew Feldman, Geoffrey Fox, Carl Kesselman,
Martin Mueller, Rick Stevens The use of AI has exploded in recent years. Many of
the applications that have been tackled make use of HPC architectures and
algorithms, as well as massive data collections. There has been much progress
and experience gained, which makes this a good time to assess what worked
well and what did not. The panelists will be asked
to identify advances in algorithms, software environments, data analytics,
standards, and computer architectures that would lead to more productive and
reliable use of AI. |
|
“The Intersection of Quantum Computing and
HPC” Wednesday, July 6th, 2022 Daniel Reed (co-chair), Christopher Monroe
(co-chair, t.b.c.) Panelists: Pete Beckman, Rick Stevens, Chad Rigetti,
Pete Shadbolt During
the past several decades, supercomputing speeds have gone from Gigaflops to
Teraflops, to Petaflops and Exaflops. As the end of
Moore’s law approaches, the HPC community is increasingly interested in
disruptive technologies that could help continue these dramatic improvements
in capability. This interactive panel will identify key technical hurdles in
advancing quantum computing to the point it becomes useful to the HPC
community. Some questions to be considered:
|
Abstracts
|
To Exascale and
Beyond: Accomplishments and Challenges for Large Scale Scientific
Visualization James
Ahrens Los Alamos National
Laboratory, Los Alamos, NM, USA Short Abstract Highlighting accomplishments from exascale visualization projects and presenting a vision
of how to support visual analysis for the evolving modern scientific process.
Long Abstract Visualization plays a critical role in the
scientific understand of the massive streams of data from scientific
simulations and experiments. Continued growth in performance and availability
of large scale supercomputing resources (e.g. exascale
soon and faster over the next decade) enables both increasing simulation
resolutions and an increasing number of and breadth of simulation ensemble
runs. In the modern scientific process these simulation ensembles are
verified for correctness and then validated with experimental ensembles to
increase our overall scientific knowledge. Effective visualization of the
verification and validation (V&V) process is a significant challenge.
Additional challenges include the significant gap between supercomputer
processing and data storage speeds. In this talk, I will highlight current
accomplishments from the U.S. Exascale Computing
Project to address these challenges include high-dimensional visual analysis,
comparative visualization, in situ visualization, portable multi-threaded
visualization algorithms, and automated techniques. I will present a vision
of a set of needed initiatives to support the visual understanding of the
complex and evolving modern scientific process. Bio Dr. James Ahrens is the director of the Information Science Technology
Institute at Los Alamos National Laboratory. He is also the Department of
Energy Exascale Computing Project (ECP) Data and
Visualization lead for seven storage, data management and visualization
projects that will be a key part of a vibrant exascale
supercomputing application and software ecosystem. His research interests
include visualization, data science and parallel computing. Dr. Ahrens is author of over 120 peer reviewed papers and
the founder/design lead of ParaView, an open-source
visualization tool designed to handle extremely large data. ParaView is broadly used for scientific visualization and
is in use at supercomputing and scientific centers
worldwide. Dr. Ahrens received his B.S. in Computer
Science from the University of Massachusetts at Amherst in 1989 and a Ph.D.
in Computer Science from the University of Washington in 1996. Dr. Ahrens is a member of the IEEE and the IEEE Computer
Society. Contact him at ahrens@lanl.gov. |
|
Trends in Parallel File Systems for HPC - A
European Perspective Frank
Baetke EOFS (European Open File
Systems - Societas Europaea) Parallel File Systems are an essential part of
almost all HPC-Systems. The need for that architectural concept originated
with the growing influence and finally complete takeover of the HPC spectrum
by parallel computers either defined as clusters or as MPPs following the
nomenclature of the TOP500. A major step towards parallel file systems for the
high end of HPC systems occurred around 2001 when the US DoE funded the
development of such an architecture called LUSTRE as part of the ASCI path
forward project with external contractors that included Cluster File Systems
Inc. (CFS), Hewlett Packard and Intel. The acquisition of the assets of CFS
by SUN Microsystems in 2007 and its subsequent acquisition by ORACLE in 2010
led to a crisis with the cancellation of future work on LUSTRE. To save the assets and ensure further development a
few HPC-focused individuals founded organizations as EOFS, OpenSFS and Whamcloud to move
LUSTRE to a community-driven development. In 2019 EOFS and OpenSFS jointly acquired the LUSTRE trademark, logo and
related assets and have jointly organized LUSTRE-focused sessions at SC and
ISC ever since. In Europe development of a parallel file system
focused on HPC began in 2005 at the German Fraunhofer
Society also as an open-source project dubbed FhGFS
(Fraunhofer Global Parallel File System) that has
now - driven by its spin-off ThinkParQ and renamed BeeGFS – gained worldwide recognition and visibility. In contrast to community-driven open-source concepts
several proprietary parallel file systems are widely in use with IBM’s
Spectrum Scale – originally known as GFPS – having the lead in HPC with a
significant number of installations at the upper ranks in the TOP500 list.
But there are other interesting proprietary concepts with specific areas of
focus and related benefits. In this talk we will review the role of EOFS
(European Open File Systems - SCE) and Europe’s focus and contribution in the
further development of this essential components of HPC-Systems. Note: All trademarks
are the property of their respective owners |
|
C is not for Cookie: Collaboration,
Convergence and Composable CI Through the Fire Lense Ilkay Altintas San Diego Supercomputer Center and Workflows for Data Science (WorDS) Center of Excellence and
WIFIRE Lab University of California at San Diego, CA, U.S.A. We are in “the age of complexity”. Our world is
increasingly influenced by grand challenge problems, requiring new integrated
systems of knowledge, technology and environment to be developed at the
societal scale. In every area of science and society, new solutions are
becoming possible through the advances in data science, information and
computing, but also through developments of collaborative teams. This talk
discusses the systems and methodology requirements such team science and
integrated applications pose on cyberinfrastructure, and provides an example
application in wildland fire science area. |
|
AI for the Instrument to HPC Continuum Pete Beckman Argonne
National Laboratory, Argonne,
IL, USA No longer does a chasm exist between scientific
instrumentation and advanced computation. From the sensor to the laptop, from
the telescope to the supercomputer, from the microscope to the database,
scientific discovery is part of a connected digital continuum that is dynamic
and fast. In this new digital continuum, Artificial intelligence (AI) is
providing tremendous breakthroughs, making data analysis and automated
responses possible across the digital continuum. SAGE is a National Science
Foundation project to build a national cyberinfrastructure for programable edge computing. The SAGE infrastructure
allows scientists to write “software-defined sensors” by analyzing
the data in situ, at the edge, at the highest resolution of data. Data from
the edge computation are then transmitted to a cloud computing infrastructure
where they can be archived and provided to the community as data products or
used in real time to trigger computational models or dynamically modify
subsequent edge computation. This new edge computing programming framework
gives scientists a new tool for exploring the impacts of global urbanization,
natural disasters such as flooding and wildfires, and climate change on
natural ecosystems and city infrastructure. SAGE is deploying
cyberinfrastructure in environmental testbeds in California, Colorado, and
Kansas, in the National Ecological Observatory Network, and in urban
environments in Illinois and Texas. Artificial intelligence will transform
the digital continuum, changing programming models and how shared scientific
facilities are designed and built. |
|
Multi-terabit photonic connectivity for
energy efficient AI computing Keren
Bergman Columbia University, USA Modern ML models now use hundreds of billions of
parameters. Such large models require distributed systems and are
increasingly bottlenecked by the energy and communications costs of
interconnection networks. Integrated silicon photonics offer the opportunity
for delivering ultra-high bandwidth connectivity that is energy efficient and
scalable system wide. We introduce the concept of embedded photonics for
deeply disaggregated architectures. Beyond alleviating the bandwidth/energy
bottlenecks, the new architectural approach enables flexible connectivity
tailored to accelerate distributed ML applications. Keren Bergman is the Charles
Batchelor Professor of Electrical Engineering at Columbia University where
she also serves as the Faculty Director of the Columbia Nano Initiative. Prof. Bergman received the B.S. from Bucknell
University in 1988, and the M.S. in 1991 and Ph.D. in 1994 from M.I.T. all in
Electrical Engineering. At Columbia, she leads the Lightwave Research
Laboratory encompassing multiple cross-disciplinary programs at the
intersection of computing and photonics. Prof.
Bergman is the recipient of the 2016 IEEE Photonics Engineering Achievement
Award and is a Fellow of Optica and IEEE. |
|
AI for Experimental Design at Urban and
Regional Scales Charlie
Catlett Discovery Partners
Institute, University of Illinois System, USA Machine learning (ML) methods, particularly for
community vulnerability assessment and forecast, have improved our ability to
optimize strategies ranging from the quantity and placement of sensors to
measure urban heat islands and air pollution to the selection of sampling points to measure
infectious disease levels in urban areas and large rural regions. The use of ML for these and other
experimental design problems will be discussed, noting the unique challenges
at different spatial scales, from census tract to city- or county-scale. |
|
Enabling HPC potential for engineering
applications in aerospace and defence Carlo Cavazzoni SVP of Cloud Computing
& Director of the HPC Lab, Chief Technology & Innovation Office,
Leonardo S.p.A., Genoa, ITALY The presentation will cover the Leonardo experience
in using its own convergent HPC and Cloud infrastructure for aerospace
applications, with particular focus on the usage of Digital Twin and BigData
technologies. |
|
Accelerating Discovery and Design using AI Henry & Isabelle Dever Professor of ECE, Northwestern
University, USA How can AI help accelerate knowledge discoveries and
exploration of design spaces. An example of this is learning from data to
build predictive models that can enable exploration of scientific questions
without relying upon underlying theory or even domain knowledge. Another
example is the acceleration of so called the “inverse problems” which explore
the design space based on desired properties. For example, can AI learn basic
chemistry from data? Or how can AI replace or reduce the need for expensive
simulations or experiments to perform discoveries quickly or evaluate a
feasible design space? This talk will present some learnings that address
some of the questions above using various materials design and discovery
examples. Biography: Dr. Alok Choudhary is the Dever
Professor of Electrical Engineering and Computer Science at Northwestern University. He also teaches at Kellogg
School of management. He is the founder, chairman and chief scientist of 4C
insights, a big data analytics and marketing technology software company (4C
was recently acquired by MediaOcean). He received
the National Science Foundation's Young Investigator Award in 1993. He is a
fellow of IEEE, ACM and AAAS. He has published more than 400 papers in
various journals and conferences and has graduated 45+ PhD students,
including more than 10 women PhDs. |
|
Trish Damkroger HPE, Germany Driven by convergence with artificial intelligence
and data analytics, increased heterogeneity, and a hybrid cloud/on-premise delivery model, dynamic composition of
workflows will be a key in designing future high-performance computing
systems. While tightly coupled HPC workloads will continue to execute on dedicated
supercomputers, other jobs will run elsewhere, including public clouds, and
at the edge. Connecting these distributed computing tasks into coherent
applications that can perform at scale is the key to harnessing the power of
HPC. |
|
16 Years Later: Pegasus in Cetraro: challenges and solutions for emerging computing
systems Ewa Deelman Los Alamos National
Laboratory, Los Alamos, NM, USA The talk will examine the challenges faced by
workflow management systems over the last two decades, what concepts survived
and what new solutions were developed to address emerging computing systems,
and where the gaps remain. In
particular the talk will focus on the Pegasus workflow management system and
its applications and describe their evolution and adaptation over time as the
execution systems have gone from tera- to exa-scale. |
|
Still having Fun After 50 Years Jack Dongarra University of Tennessee,
Oak Ridge National Laboratory, University of Manchester, USA & United
Kingdom In this talk, we will look back at some of the
highlights I have had good luck in being involved in. |
|
Exploration of QUANTUM COMPUTING @ Leonardo
Labs A pragmatic industrial approach Daniele Dragoni Leonardo S.p.A., High
Performance Computing Lab, Genova, Italy Quantum Computing (QC) is an emerging paradigm that
offers the potential to solve complex problems that are considered
intractable within the classical/digital computing domain. Building a quantum
computer capable of solving problems of practical interest is, however, an
engineering challenge unmatched. In fact, despite the rapid pace of
development of the hardware technologies that has recently enabled the first
demonstrations of quantum supremacy, no evidence of quantum advantage on
real-world problems has been observed yet. Nonetheless, we are currently
witnessing an era of quantum enablement where industries, attracted by
glimpses of quantum computational capabilities, have started to investigate
potential benefits associated with this technology. In this talk, I will present the approach Leonardo
is taking to explore the QC domain, introducing the research areas and applications
that we intend to investigate by means of emulators on HPC systems and
quantum hardware. Finally, I will show some examples of activities that we
are carrying out within these research areas. |
|
Powering the Future: AI, AI+HPC at Wafer
Scale Andrew
Feldman Founder and CEO Cerebras Systems, Sunnyvale, California, USA Artificial intelligence (AI) has shown
transformative potential for scientific discovery in applications ranging
from life sciences and medicine to physics and energy. In addition, AI is
being used alongside or in conjunction with traditional high performance
computing (HPC) routines to build better and more efficient simulation and
processing pipelines. While the potential of this work is high, the
associated compute demands are large and growing much faster than Moore’s law
allows us to build processors. Large
scale AI, HPC, and AI+HPC for science requires massive sparse compute, near
compute memory and high bandwidth communication – attributes not found in
clusters of traditional processors. In this talk, Cerebras
will describe its wafers scale processor, the WSE-2 and the software
architecture that takes advantage of the 850,000 near memory compute cores.
We will discuss the chip architecture, hardware software co-design, and the
challenges presented to engineering by wafer scale. We will show real world use cases in AI,
HPC, and AI+HPC that show unique advantage in scientific and commercial
applications and we will also present new research enabled by the WSE-2’s
architecture for training very large models with high levels of sparsity. |
|
Intelligent Simulations: How Combining AI and
HPC Can Enable New Discoveries Ian
Foster Argonne National
Laboratory, USA The search for ever-more accurate and detailed
simulations of physical phenomenon has driven decades of improvements in both
supercomputer architecture and computational methods. It seems increasingly
likely that the next several orders of magnitude improvements are likely to
come, at least in part, from the use of machine learning and artificial
intelligence methods to learn approximations to complex functions and to
assist in navigating complex search spaces. Without any aspiration for
completeness, I will review some relevant activities in this space and
suggest some implications for future research. |
|
Deep Learning for Earthquake Nowcasting Geoffrey
Fox Indiana University, USA AI is expected to transform both science and the
approach to science. As an example, we take the use of deep learning to
describe geospatial time series and present a general approach building on
previous work on recurrent neural networks and transformers. We give examples
of so-called spatial bags from earthquake nowcasting,
hydrology, medical time series, and particle dynamics and focus on the
earthquake case. The latter is presented as an MLCommons
benchmark challenge with three different implementations: a pure recurrent
network, a Spatio-temporal science transformer, and
a version of the Google Temporal Fusion Transformer. We discuss the physics
intuition that hidden variables may be elusive for Earthquakes but more
natural in for example hydrology. We discuss
deep learning issues such as the seemingly unreasonable number of
parameters (weights), software engineering implications, and the need for
significant computational resources Reference https://www.mdpi.com/2624-795X/3/2/11/htm |
|
Enhancing the Engineering Simulation HPC
Cloud Platform and HPC Containers Towards Multi-Cloud and Kubernetes Wolfgang
Gentzsch and Daniel Gruber The UberCloud,
Sunnyvale, California, and Regensburg, Germany Over the past 12 months, we extended our cloud
application platform and the HPC containers towards several directions.
First, early last year, after it has been announced that CentOS will be
‘end-of-lifed’, we entered into a technology
partnership with SUSE and jointly ported our containers from CentOS to SUSE
Linux Enterprise Server (SLES) and to the SUSE Base Container Image (BCI) and
tested it successfully with several commercial application codes. Second, we further improved our multi-cloud
engineering simulation platform. It is now used in production on Azure,
Google Cloud, and AWS, and we used it in our Living Heart Project with Dassault Systèmes and 3DT
Holdings which has been widely recognized in the HPC community. And third, we enhanced our platform with new
Kubernetes cluster management capabilities for environments like AWS (EKS),
Azure (AKS), Google (GKE), and SUSE (RKE). Each of these managed Kubernetes
environments has their specialties and challenges that we successfully
tackled one by one. Finally, we tested / benchmarked these environments that
showed different performances. |
|
Component-Based Message Passing for Java: Developments,
Achievements, and Impact Vladimir
Getov Distributed and
Intelligent Systems Research Group, University of Westminster, London, U.K. Released in 1995, the Java programming language was
rapidly adopted by industry and end users because of its portability and
internet programming support. However, Java did not have the symmetric
message passing capability, widely recognised as vitally important for
parallel and distributed memory computing. By contrast, efficient message
passing support had already been captured in the MPI (message-passing
interface) standard for other programming languages such as C, C++, and
Fortran. To alleviate this difficulty, various early projects including our
own work started the development of Java message-passing systems. Then, a
single MPI-like API specification and reference implementation were developed
by the Message Passing for Java (MPJ) International Working Group as part of
the JavaGrande Forum. This group also developed a
methodology for building mixed-language MPJ applications which evolved from
three approaches: (a) wrapping of existing MPI libraries via hand-written
software; (b) automatic code generation of the wrapper interface by a novel
tool-generator; and (c) development from scratch of the MPJ libraries in
Java. The development of all three approaches implemented the MPJ
specification which successfully resembled MPI, providing symmetric message
passing for distributed computing with Java. Nowadays, MPJ is the predecessor of the Java binding
included since 2014 in the core distribution of the widely used Open MPI
software environment. The invention of MPJ resulted in an industry standard
specification and a novel component-based hierarchical development
methodology which enables the development of very large and complex grid and
cloud-based distributed systems. These achievements led to: (a) impact on
professional practice and development productivity; (b) significant economic
impact; and (c) social impact via the results of the novel component-based application. |
|
Unlocking the Power of Quantum with D-Wave Victoria
Goliber D-Wave Systems Inc.,
Germany & USA As quantum technologies advance, more and more
customers are seeing the value in bringing quantum into their business. With
D-Wave’s quantum annealers and hybrid
quantum-classical solvers, we now have the ability to solve real-world
problems with up to one million variables. In addition, upgraded software
tools provide easy-to-use functionality for users to quickly translate their
business problems into quantum applications. Join us to hear about the latest
releases and industry case studies. |
|
Innovating and democratizing HPC, challenges for
a platform vendor Martin Hilgeman Dell Technologies, USA It is the Dell Technologies’ mission to make HPC
systems available to everyone, with an emphasis on ease of use, standards
compliance without vendor lock-in, while also advancing HPC through research.
Most of Dell’s HPC research is done at the HPC and AI Innovation Lab, which
is hosted on the Dell campus in Austin, TX. This presentation gives an
overview of the lab’s capabilities, amended with selected case studies. The
author also discusses the challenges that HPC platform vendors like Dell
Technologies face in terms of enabling application efficiency, while using
massively parallel and multi-core processors, domain specific accelerators,
and large-scale parallel storage systems. Biography Martin joined Dell Technologies in 2011,
after having worked as an HPC application specialist for 12 years at SGI and
IBM. In 2019, he joined AMD as a senior manager and worked on porting and
optimizing the major HPC applications to the “Rome” microarchitecture. Martin
returned to Dell Technologies in May 2020 as the HPC performance lead and
Distinguished Member of Technical Staff in Dell ISG. He owns a master’s
degree in physical chemistry, obtained at the VU University of Amsterdam. |
|
Enabling quantum-classical computation on
multiple time scales Blake
Johnson Distinguished RS,
Quantum Platform Lead, IBM Quantum It is increasingly apparent that quantum computers
will not enter the world of computing as standalone entities. Rather, they
will be used in concert with classical computers to solve problems. These
interactions take several forms and occur on several distinct time scales,
from the ultra-low latency domain of dynamic circuits utilizing feedforward of
quantum measurements, to the domain of scalable elastic compute with cloud
HPC. These considerations motivate new interfaces to quantum hardware to
express such interactions and/or enable integration in other compute models. I will discuss developments in two of
these domains, including the OpenQASM3 circuit description language for
combining real-time classical computation with quantum operations, and the Qiskit Runtime that powers higher-level primitives that
allow for managed performance of algorithmic building blocks. Time
permitting, I will provide an outlook for how elastic cloud HPC might be
combined with Runtime primitives to extend the computational power of quantum
systems. |
|
R&D of a Quantum-Classical HPC Hybrid
Platform and its Target Applications with QA-Based Simulation-Data Analysis
Hybrid Architecture Laboratory,
Department of Computer and Mathematical Sciences, Graduate School of
information Sciences, Tohoku University, JAPAN In this talk, I will be presenting our on-going
project entitled Quantum-Annealing Assisted Next-Generation HPC
Infrastructure, In this project, we try to realize transparent
accesses to not only classical HPC resources with heterogeneous computing
platforms such as x86 and vector accelerator, but also Quantum Computing one
in a unified fashion. We are also developing next-generation applications
in the fields of computational science, data sciences and their fusions best
suited for this infrastructure. The target applications are three digital twins:
Digital Twin for Disaster Resilience, Digital Twin for Stable Power
Generation and Digital Twin for Soft-Material Design. In these applications developments, we introduce
quantum annealing and simulated annealing accelerated by classical HPC into
optimal evacuation route analysis and data clustering, respectively. Some performance discussion on different type of annealers by quantum and classical computing is also
presented in this talk. |
|
LUMI EuroHPC
architecture and impact for research Kimmo Koski CSC - Finnish IT Center for Science, Espoo, Finland The #1 European supercomputer LUMI hosted by CSC at the Kajaani
data centre in Finland is an 200 MEUR
investment by 10 European countries and the European Commission. A peak flops
GPU partition is complemented not only by a x86 based partition, but also
with different large storage capabilities in form of fast flash-based disks, large parallel file
system and object store for sharing and staging data. A cloud container
partition to support complex workflows and a partition for interactive data
analytics and visualization are included. One additional feature of LUMI is that it from beginning includes a
support for quantum computing. Initially by a quantum simulator/emulator and,
as presented in this, direct access to several quantum computers. By this way
making experimentation with quantum computing available for the advanced HPC
users of LUMI our aim is to accelerate the up-take of quantum computing
within the computational sciences, stimulate co-creation with the quantum
computing community and, contribute to development of the quantum computing
and HPC integration software stack. The architecture of the installation is aimed to
satisfy a broad range of advanced research needs. Having LUMI installed and
operational during this summer in its full capability, over half-an-exaflop performance, is only the beginning of the story.
The next steps include providing world class computing capability for various
applications and research projects, targeting to address grand challenges of
computing in different fields – number of which are discussed in this talk. |
|
What we Learned About Programming Models
During the Race to Exascale Stefano Markidis Computer Science
Department, KTH Royal Institute of Technology, Stockholm, Sweden In late May 2022, the Frontiers supercomputer at the
Oakridge national laboratory broke the exaflop
barrier ending the international race to deliver the first exaflop supercomputer. The significant engineering effort
in designing, developing, and integrating hardware for the exascale machine has been accompanied by substantial
developments within programming models. Traditional programming approaches
further developed to keep up with and ease the programmability of extremely
heterogeneous compute and memory systems; new programming approaches emerged and
established themselves as credible HPC programming systems. In this talk, I
review the recent developments in programming models and discuss the current
challenges to improve them in the post-exascale era
further. |
|
Towards FugakuNEXT
- Life after Exascale Satoshi
Matsuoka Tokyo Institute of
Technology, Japan Fugaku, the Japanese
flagship supercomputer commissioned early 2021 at Riken R-CCS, has been a great
success as the first supercomputer to demonstrate exascale
performance across a variety of applications pertaining to the Japanese
Society 5.0 i.e., digital transformations towards solving the society’s most
important problems to realize SDGs. Now, we are embarked on a possible path
towards FugakuNEXT, which will be a successor to Fugaku, to be launched towards the end of this decade.
The goal is to significantly amplifying the two key metrics of Fugaku, namely high performance and broad applicability,
but of course such would be even harder to achieve compared to Fugaku, especially with semiconductor advances slowing
down. The talk will introduce the current efforts towards the feasibility
study which will commence August 2021, but already, some ideas have been
explored with our recent set of research, which gives us good guidance
towards a novel, somewhat different trajectory than the exascale
machines today. |
|
Elena
Messina Principal, Prospicience LLC. USA Artificial intelligence (AI) technologies and
robotic systems have been intertwined since their inceptions. Recently, there
has been an explosion of research and development in applying machine
learning (ML) to robotic perception, planning, grasping, and human
interaction, enabled by broad availability of datasets and greater
computational resources. This new generation of learning algorithms relies on
vast quantities of data, which is currently obtained via ad hoc procedures,
typically without traceability, quality measures, or ways to characterize
applicability. Expectations for the advancements in robotics that ML and
other AI technologies will enable are quite high. Examples of anticipated
breakthroughs include autonomous vehicles used as transportation and for
deliveries, home robots that allow aging in place, and highly dexterous and
adaptive robots that can plan and perform manufacturing operations as well as
humans. Are these expectations grounded in reality? The great
promise of ML-augmented robotics raises the need for a measurement science
infrastructure to help assess the technology’s maturity and guide researchers
towards highly-reliable solutions. Machine learning algorithms are only as good as the
data that is used to train them. Therefore the datasets that are used must be
vetted and characterized. Additionally, the applicability and limitations of
the resulting systems need to be well-understood in order to have effective,
safe, and reliable deployments. Robots are an example of embodied AI. Their
physicality presents additional challenges in training and execution due to
unpredictable interactions with humans and real-world objects and
environments. In this talk, I will discuss some of these challenges as well
as efforts that are underway at the U. S. National Institute of Standards and
Technology and other organizations to address the missing measurement science
to help guide researchers, practitioners, and policy makers. |
|
Christopher
Monroe Duke University and IonQ, Inc., USA Trapped atomic ions are a leading physical platform
for quantum computers, featuring qubits with essentially infinite idle
coherence times and the highest purity quantum gate operations. Such atomic
clock qubits are controlled with laser beams, allowing densely-connected and
reconfigurable universal gate sets. The path to scale involves concrete
architectural paths, from shuttling ions between QPU cores to modular
photonic interconnects between multiple QPUs. Full-stack ion trap quantum
computers have thus moved away from the physics of qubits and gates and
toward the engineering of optical control signals, quantum gate compilation
for algorithms, and software-defined error correction. I will summarize the
state-of-the-art in these quantum computers in both academic and industrial
settings, and summarize how they are being used for both scientific and
commercial applications. |
|
Reconfigurable Dataflow Architecture - AI
made easy Martin
Mueller Sambanova Systems, Germany Sambanova Systems developed
a novel approach to process neural-network like AI challenges of nearly
arbitrary size. The talk will introduce you to the company, the
“Reconfigurable Dataflow Architecture”, and the components that make up the
whole platform. Speaker bio: Martin has 25yrs of experience
in various different technical roles in information technology, from being
part of microprocessor development group, field technical presales to product
management roles. He holds a diploma in theoretical physics, and lives in
Germany. |
|
Benchmarking and Analysis of Noisy Intermediate-Scale
Trapped Ion Quantum Computing Architectures Kevin Obenland Quantum Information and
Integrated Nanosystems, Lincoln Laboratory,
Massachusetts Institute of Technology MIT, USA Quantum computing is at the cusp of showing relevance for real-world
problems. However, currently available devices are small in scale, and
hardware demonstrations have shown limited applicability to scientific and
commercial problems of interest. In this work, we investigate a set of
application-specific noisy intermediate-scale quantum (NISQ) algorithms
varying in size from 4-80 qubits, and use these benchmarks to evaluate
trade-offs in the design of 5 candidate trapped ion quantum computing
architectures. We have developed a tool-chain consisting of architecture specific
compilation and simulation tools, which allows us to estimate metrics such as
application run-times and overhead. Additionally, we use our tools to
determine the critical operations of an architecture and to study the
sensitivity to architectural constraints in particular implementations. Our
tools are designed to be flexible, allowing us to study a broad range of
benchmarks, hardware architectures, physical constraints, and operation
timing. |
|
Towards the Intelligent Discovery and
Delivery of Major Facility Data Manish Parashar Rutgers University, USA Data collected by large-scale instruments,
observatories, and sensor networks, i.e., science facilities, are key
enablers of scientific discoveries in many disciplines. However, ensuring
that these data can be accessed, integrated, and analyzed
in a democratized and timely manner remains a challenge. In this talk, I will
explore how state-of-the-art techniques for data discovery and access can be adapted
to facility data and develop a conceptual framework for intelligent data
access and discovery. |
|
The National Science Data Fabric: Democratizing Data Access for Science and
Society John R. Parks Endowed
Chair, University of Utah, Professor, School of Computing, Faculty,
Scientific Computing, and Imaging Institute,
Director, Center for Extreme Data
Management, Analysis and Visualization (CEDMAV), USA Effective use of data management techniques for the
analysis and visualization of massive scientific data is a crucial ingredient
for the success of any experimental facility, supercomputing center, or cyberinfrastructure that supports
data-intensive scientific investigations. Data movements have become a
central component that can enable or stifle innovation in the progress
towards high-resolution experimental data acquisition (e.g., APS, SLAC, NSLS
II). However, universal data delivery remains elusive, limiting the
scientific impacts of these facilities. This is particularly true for
high-volume/high-velocity datasets and resource-constrained institutions. This talk will present the National Science Data
Fabric (NSDF) testbed, which introduces a novel trans-disciplinary data fabric
integrating access to and use of shared storage, networking, computing, and
educational resources. The NSDF technology addresses the key data management
challenges involved in constructing complex streaming workflows that take
advantage of data processing opportunities that may arise while data is in motion. This technology finds
practical use in many research and industrial applications, including
materials science, precision agriculture, ecology, climate modeling, astronomy, connectomics,
and telemedicine. This NSDF overview will include several techniques
that allow building a scalable data movement infrastructure for fast I/O
while organizing the data in a way that makes it immediately accessible for
processing, analytics, and visualization with resources from Campus Computing
Cybeinfrastructures, the Open Storage Network, the
Open Science Grid, NSF/DOE leadership computing facilities, the CloudLab, Camelion, and
Jetstream, just to name a few. For example, I will present a use case for the
real-time data acquisition from an Advanced Photon Source (APS) beamline to
allow remote users to monitor the progress of an experiment and direct
integration in the Materials Commons community repository. We accomplish this
with an ephemeral NSDF installation that can be instantiated via Docker or
Singularity at the beginning of the experiment and removed right after. In
general, the advanced use of containerized applications with automated
deployment and scaling makes the practical use of clients, servers, and data
repositories straightforward in practice, even for non-expert users. Full
integration with Python scripting facilitates the use of external libraries
for data processing. For example, the scan of a 3D metallic foam can be
easily distributed with the following Jupyter
notebook https://bit.ly/NSDF-example01. Overall, this leads to building flexible data
streaming workflows for massive imaging models without compromising
the interactive nature of the exploratory process, the most effective
characteristic of discovery activities in science and engineering. The
presentation will be combined with a few live demonstrations of the same
technology including notebooks which are being used to provide undergraduate
students of a minority-serving institution (UTEP) with real-time access to
large-scale data normally used only by established scientists in well-funded
research groups. BIOGRAPHY Valerio Pascucci is the Inaugural John R. Parks Endowed Chair, the
founding Director of the Center for Extreme Data
Management Analysis and Visualization (CEDMAV), a Faculty of the Scientific
Computing and Imaging Institute, and a Professor of the School of Computing
of the University of Utah. Valerio is also the President of ViSOAR LLC, a University of Utah spin-off, and the founder
of Data Intensive Science, a 501(c) nonprofit
providing outreach and training to promote the use of advanced technologies
for science and engineering. Before joining the University of Utah, Valerio
was the Data Analysis Group Leader of the Center
for Applied Scientific Computing at Lawrence Livermore National Laboratory
and an Adjunct Professor of Computer Science at the University of California,
Davis. Valerio's research interests include Big Data management and
analytics, progressive multi-resolution techniques in scientific
visualization, discrete topology, and compression. Valerio is the coauthor of more than two hundred refereed journal and
conference papers and was an Associate Editor of the IEEE Transactions on
Visualization and Computer Graphics. |
|
Atom Computing’s Quantum Platform Kristen Pudenz Atom Computing,
Berkeley, CA, USA Atom Computing is creating a quantum processing platform
based on nuclear spin qubits. The system makes use of optical tweezers to
assemble and individually manipulate a two-dimensional register. We will
explore progress on the Phoenix hardware platform and the potential of the
technology to create scalable quantum computing solutions. |
|
Reinventing High-Performance Computing Daniel
Reed University of Utah, USA Although there are exceptions, most HPC systems on
the TOP500 are examples of a commodity monoculture, built from nodes
containing server-class microprocessors and the same GPU accelerators widely
used for machine learning. With the end of Dennard scaling, the slowing of
Moore’s Law, and exponentially rising costs for semiconductor fabrication facilities,
high-performance computing (HPC) is at an important inflection point, with
deep and profound change now in the wind. In addition to the technical
challenges of chip design, the global semiconductor shortage and associated
political battles surrounding fabrication facilities are now affecting
everyone. In another profound shift, computing economics are now dominated by
cloud hyperscalers and smartphone vendors who are
increasingly building using custom semiconductors. |
|
Path To Large Scale Fault-Tolerant Quantum
Computing Chris Rigetti Founder and CEO Rigetti Computing, Berkeley, CA, U.S.A. Over the past decade gate-model quantum computers
with ~100 qubits have gone from concept to commercial reality. Current
practical applications include quantum machine learning, optimization, and
simulation and center on the use of hybrid
quantum/classical algorithms as the core computational subroutines. Future
systems aim to reduce physical error rates to increase computational utility,
and to deliver vastly larger numbers of qubits with these low error rates to
enable error correction and fault-tolerant routines. |
|
Quantum computing with neutral atom qubits Mark Saffman ColdQuanta and University of
Wisconsin-Madison, USA One of the daunting challenges in developing a
computer with quantum advantage is the need to scale to a very large number
of qubits while maintaining the fidelity and isolation of pristine, few qubit
demonstrations. Neutral atoms are one of the most promising approaches for
meeting this challenge, in part due to the combination of excellent isolation
from the environment and the capability to turn on strong two-qubit
interactions by excitation to Rydberg states. I will present our recent
results on running quantum algorithms on a neutral atom processor. Progress
towards scaling up the array, high fidelity gate operations, and mid-circuit
measurements for error correction, will also be described. |
|
What accelerated data analytics and computing
applied to numerical weather prediction is telling us about the future of
supercomputing Thomas
Schulthess CSCS Swiss National
Supercomputing Centre, Lugano and ETH, Zurich, SWITZERLAND Numerical Weather Prediction (NWP) was one of the
new domains that emerged in the early days of electronic computing and
continues to play a central role in supercomputing today. For almost a decade
CSCS and its partners at MeteoSwiss, as well as at
the Center for Climate Systems Modelling and the
Scalable Parallel Computing Lab of ETH Zurich, have exploited the virtues of
accelerated computing and data analytics for NWP and climate modelling. I
will share these experiences, as well as the conclusions we have drawn for
the design of the “Alps” infrastructure, with which CSCS will be offering
future extreme computing and data solutions to support the evolving needs of
science. |
|
Silicon photonic quantum computing Pete Shadbolt Chief Scientific Officer
& Co-Founder PsiQuantum, Palo Alto, CA, USA It is increasingly accepted that all commercially
useful quantum computing applications require error-correction and therefore
at least 1 million physical qubits. The manufacturing capability and
expertise of the semiconductor industry are needed to deliver a commercially
useful quantum computer on any reasonable time or money scale. In this talk,
we will show how unique technology in the areas of silicon photonics and
quantum system architecture enable the path to manufacturability and
scalability of a fault-tolerant, general-purpose 1-milliion qubit quantum
computer. |
|
Cloud-Native Supercomputing: Bare-Metal,
Secured Supercomputing Architecture Gilad Shainer NVIDIA, USA High-performance computing and artificial
intelligence have evolved to be the primary data processing engines for wide
commercial use, hosting a variety of users and applications. While providing
the highest performance, supercomputers must also offer multi-tenancy
security. Therefore they need to be designed as cloud-native platforms. The
key element that enables this architecture is the data processing unit (DPU).
DPU is a fully integrated data-center-on-a-chip
platform that can manage the data center operating
system instead of the host processor, enabling security and orchestration of
the supercomputer. This architecture enables supercomputing platforms to
deliver bare-metal performance, while natively supporting multi-node tenant
isolation. We’ll introduce the new supercomputing architecture, and include
applications performance results. |
|
Transition to Memory-Centric HPC Architecture
for Data Analytics Thomas
Sterling Indiana University,
Bloomington, USA Matrix-based applications exhibit memory access
patterns that favor temporal and spatial locality, allowing
effective use of caches and registers to mitigate performance degradation due
to latency effects. Such conventional methods also make better use of memory
bandwidth; both using variations of conventional von Neumann core
architectures. Two recent trends in the domain of HPC-AI are now constraining
future advances in performance progress. One is the flat-lining of Moore’s
Law and the other is the demands of data-intensive analytics and AI
applications. To address both of these barriers, a new generation of
memory-centric computer architectures are being explored as a means of
accelerating time-varying graph computation or even serving in an innovative
highly scalable standalone data-oriented computing platforms. This
presentation will describe the Pilot-CRIS experimental memory-centric
architecture under investigation, in part by Indiana University, to address
both of these strategic challenges for future AI. Pilot-CRIS is one of a
number of leading-edge architectures under development or recently deployed
to exercise these new opportunities from academic and start-up companies to
some of the biggest corporations in the world. The principles of Pilot-CRIS
and their potential will be described in detail. Pilot-CRIS is to be
sponsored by the ARO and IARPA AGILE research program. |
|
Trillion Parameter Language Models
pre-trained with Science Text and Data: Is this a plausible path towards the
development of Artificial General Intelligence for Science? Rick
Stevens Argonne National
Laboratory and University of Chicago, USA In the last three years study after study has
demonstrated the unexpected power and flexibility of large-scale Generative Pre-trained
Transformer 3 (GPT-3) like Large Language Models (LLMs) with >>100
billion parameters pre-trained on >>100 billions
of input tokens. These state-of-the-art LLMs need exascale
class machines for training and development and may represent a major
emerging class of applications for exascale
systems. LLMs can be used in a variety of applications, such as answering
questions, summarizing documents, translating, planning, writing programs,
generating step-by-step directions for common laboratory procedures,
associating the function of genes and the mechanism of action of drugs and
many more uses. Unlike common narrow problem specific deep learning models
widely used in scientific and commercial domains and generally trained in a
supervised fashion, LLMs are pre-trained in an unsupervised fashion on
large-scale collections of general text, images, and code. Our group and others are investigating how
to assess the performance of existing LLMs in the context scientific and
technical tasks, as well studying how performance on scientific tasks might
be improved by augmenting general knowledge training datasets with 10’s of
millions of scientific papers, scientific images, and structured scientific
datasets from domains such as biology, chemistry, and materials science. In this talk I’ll review recent progress in
building general purpose LLMs that may be useful for downstream scientific
tasks. I’ll review the approach to building, training and testing LLMs,
including computing needs and our emerging multi-lab and University
collaborative project of developing a trillion-parameter general purpose
“foundation” model based on LLM that we can adapt for scientific data
integration, data distillation, scientific tasks in design and analysis and
scientific hypothesis formation. It
seems clear that trillion parameter models are important and maybe a
significant stepping-stone on the path towards Artificial General
Intelligence for science. |
|
HPC and Machine Learning for Molecular
Biology: the ADMIRRAL Project* Fred Streitz Center for Forecasting and
Outbreak Analytics (CFA/CDC) and Lawrence Livermore National Laboratory
(LLNL/DOE) U.S.A. The combination of high performance computing (HPC)
and Machine Learning (ML) has proven to be a fruitful one, as evidenced by
the number of scientific disciplines that have seen advances through their
joint application. One of the most powerful demonstrations has been in the
area of computational biology, where the addition of ML techniques has helped
ameliorate the lack of clear mechanistic models and often poor statistics
which has impeded progress in our understanding. I will discuss the
development of a hybrid ML/HPC approach to investigate the behavior of an oncogenic protein on cellular membranes in
the context of the ADMIRRAL (AI-Driven Machine-learned Investigation of
RAS-RAF Activation Lifecycle) Project, a collaboration between the US
Department of Energy and the National Cancer Institute. *This work was performed under the auspices
of the U.S. Department of Energy (DOE) by Lawrence Livermore National
Laboratory (LLNL) under Contract DE-AC52-07NA27344 and under the auspices of
the National Cancer Institute (NCI) by Frederick National Laboratory for Cancer
Research (FNLCR) under Contract 75N91019D00024. This work has been supported
by the Joint Design of Advanced Computing Solutions for Cancer (JDACS4C)
program established by the U.S. DOE and the NCI of the National Institutes of
Health. |
|
HPC Acceleration of Progress in Fusion Energy
Prediction & Control Enabled by AI/Deep Learning William
Tang Princeton University,
USA This presentation discusses recent HPC advances in
the rapidly growing application of AI/Deep Learning to accelerate progress in
scientific discovery for Fusion Energy Science. Attention is focused on
constraints encountered for existing and expected algorithms and data
representations when dealing with the major challenge of prediction and
control in magnetically-confined thermonuclear plasmas relevant to the
international $25B ITER burning plasma. Associated techniques have enabled
new avenues of data-driven discovery in this quest to deliver clean energy as
one of the most prominent grand challenges for the world today. |
|
The Curios Case of Reproducing Scientific
Results about Black Holes Michela Taufer University of Delaware,
USA In 2016, the LIGO Collaboration announced the first
observation of gravitational waves from a binary black hole merger, known as
GW150914. In 2019, the Event Horizon Telescope (EHT) Collaboration announced the generation of the first image
of a black hole, called M87. Both collaborations used large-scale scientific
data and workflows to measure significant astrophysical events.
Reproducibility remains a challenging aspect of these and other large-scale
scientific workflows: reproducibility can be limited by the availability of
data, software, platforms, and documentation. To enable the understanding,
analysis, and use of the GW150914 and M87 published scientific results, we
developed sustainable knowledge necessary to reproduce the projects’ outcome.
In this talk, we discuss the challenges we encountered and we share
recommendations to make large-scale scientific workflows reproducible.
Specifically, we leverage the two projects to investigate the impact of data
availability and integrity. Furthermore, we model the GW150914 and M87
workflows and study limitations in terms of software availability,
dependencies, configuration, portability, and documentation. Our work can
enhance the reproducibility of scientific projects such as the LIGO and EHT,
empowering the scientific community (i.e., including postdocs and students,
regardless of the domain) to address similar challenges for other projects. Bio: Michela Taufer
is an ACM Distinguished Scientist and holds the Jack Dongarra
Professorship in High Performance Computing in the Department of Electrical
Engineering and Computer Science at the University of Tennessee Knoxville
(UTK). She earned her undergraduate degrees in Computer Engineering from the
University of Padova (Italy) and her doctoral
degree in Computer Science from the Swiss Federal Institute of Technology or
ETH (Switzerland). From 2003 to 2004 she was a La Jolla Interfaces in Science
Training Program (LJIS) Postdoctoral Fellow at the University of California
San Diego (UCSD) and The Scripps Research Institute (TSRI), where she worked
on interdisciplinary projects in computer systems and computational
chemistry. Michela has a long history of
interdisciplinary work with scientists. Her research interests include
scientific applications on heterogeneous platforms (i.e., multi-core
platforms and accelerators); performance analysis, modeling
and optimization; Artificial Intelligence (AI) for cyberinfrastructures (CI);
AI integration into scientific workflows, computer simulations, and data
analytics. She has been serving as the principal investigator of several NSF
collaborative projects. She also has significant experience in mentoring a
diverse population of students on interdisciplinary research. Michela’s
training expertise includes efforts to spread high-performance computing
participation in undergraduate education and research as well as efforts to
increase the interest and participation of diverse populations in
interdisciplinary studies. |
|
HPC and Sustainability: The Smarter Path to
Zero Emission Computing Scott
Tease Vice President HPC&
AI Lenovo Super Computing tackles some of the most challenging
tasks on the planet. The work we do in weather forecasting and climate
modelling/simulation is helping us understand what the future of our planet
looks like and what must be done today to avoid a crisis. We as an HPC
community know that it is essential to drive research in a more sustainable
way that protects the climate as while we drive the research needed to solve
humanity’s greatest challenges. This talk will discuss steps we can take to
drive out carbon emissions and lighten our environmental impact as we deploy,
use and dispose of IT. We can make a difference today – let’s talk about how
to start. |
|
The road to Zettascale
from an application perspective, and a few other details Philippe
Thierry Sr Principal Engineer,
Intel, France As the exascale boundary
has recently been crossed, it is time to come back to the question « How can
we achieve another 1000x » to reach a ZettaFlops,
at least while running a simple benchmark? If we assume that the size of computing centers
remains roughly constant and that we cannot really increase the energy
envelope, or we better decrease it, we
fall back on a
performance/density problem for
all levels that constitute a computer. The computing unit today is able to reach about 100
TF/s in 1 kW (dp64) ; to reach 100 PF/s in a few years, that translates to
100 Roadrunner (#1 in the top500 of June 2008) or 1 Summit (June 2018) in a
socket. We’ve proven over the years that we can meet and
surpass large computing challenges, so we can hold hope; but knowing the real usage of these machines today,
will we be lucky enough to manage to run applications without excessive
efficiency loss? If we consider that the applications and the
associated test cases can theoretically scale by simple oversampling (a
factor of 10 in the 3 dimensions) and assuming that the numerical schemes and
the physics can remain stable, then the problem comes back again to the
conservation of the ratios of bandwidth, floating point (and other data
types), latencies for scalability of applications, and reduction of the
energy needed to move a bit. It is difficult to predict if the current model of
building HPC Supercomputers can be
scaled without new breakthroughs, or if limits will be reached. A few Angstrom separate us from the size of an atom
to make a transistor. So what else?
Silicon photonics, quantum computing, cooling are words that come up
frequently as methods to address limits. In this presentation, we will briefly discuss these
points as well as the upcoming initiatives and investments in Europe to put applications
and software at the center of the problem when
designing new technologies. Additionally we will focus on performance and power
prediction from the SOC to the complete system from an application
perspective. |
|
Innovating the Next Discontinuity Robert
Wisniewski HPC Architecture Chief,
Samsung Electronics, and Samsung’s SAIT Systems, Architecture Lab Head,
U.S.A. A growing number of classical HPC applications - modeling and simulation applications - are bottlenecked
due to insufficient memory bandwidth. At the same time, AI applications,
which are forming an increasingly important part of HPC, and compute in
general, are often bottlenecked because of insufficient communication (node
to node) bandwidth. In addition, the ability to leverage efficient
accelerator cycles for both types of applications is key towards continuing
the exponential growth for post-exascale computing.
In this talk I will describe the key trends identified above, and discuss the
research we are undertaking to design the hardware and software architecture
for HPC and AI applications to obtain the next level of exponential increase
in performance. I will suggest a path forward based on leveraging tightly
integrating memory and compute, and describe the interesting design space
that needs to be considered to make this architecture a reality. This
capability has the potential to be the next discontinuity in HPC and AI. |