Computing Sciences at Berkeley Lab CS Student Program Welcome - PowerPoint PPT Presentation

Computing Sciences at Berkeley Lab CS Student Program Welcome 2 June 2020 David Brown, Director Computational Research Division Lawrence Berkeley National Laboratory

Why the nation needs national laboratories § Discovery science § Scientific solutions addressing national challenges, especially energy § Unique scientific capabilities § User facilities § Managed, large research teams § Important technologies with long, risky R&D paths § A diverse group of highly trained, creative, and committed scientists and engineers.

Berkeley Lab is one of the 17 U.S. Department of Energy (DOE) National Laboratories The mission of the Energy Department is to ensure America’s security and prosperity by addressing its energy, environmental and nuclear challenges through transformative science and technology solutions.

Berkeley Lab Changes Science Radiation Lab staff on the magnet yoke for the 60-inch Today, Berkeley Lab has: cyclotron, 1939, including: Over 4000 employees E. O. Lawrence $1.1B in FY18 funding Edwin McMillan 13 associated Nobel Glenn Seaborg prizes Luis Alvarez J. Robert Oppenheimer Robert R. Wilson

Berkeley Lab brings Science Solutions to the World Turned Confirmed the Explained Revealed the Discovered 16 Unmasked a Identified Fabricated the windows into Big Bang and Photosynthesis secrets of the elements dinosaur killer good and bad smallest energy savers discovered dark human genome cholesterol machines energy https://www.lbl.gov/program/35-breakthroughs/

Molecular Foundry ALS-Advanced Light Source Bldg 50 Wang Hall Bldg 59

Computing Sciences at Berkeley Lab in 2020 Jonathan Carter Computational Research NERSC David Brown Computational Sudip Dosanjh Science Applied Mathematics Scientific Networking: ESnet Computer Inder Monga Science Data Science & Technology 10

NERSC at Berkeley Lab Provides HPC and Data Resources for Science Research Largest funder of physical science research in U.S. Materials, Chemistry, Biology, Environment Computing Geophysics Particle Physics, Nuclear Physics Fusion Energy, Astrophysics Plasma Physics

NERSC’s newest machine Cori supports both the HPC Workload and Data-Intensive Science • Cray system with 9,300 Intel Knights Landing compute nodes – Self-hosted, (not an accelerator) manycore processor > 64 cores per node – On-package high-bandwidth memory at >400GB/sec • Data Intensive Science Support – 10 Haswell processor cabinets (Phase 1) to support data intensive applications – NVRAM Burst Buffer with 1.5PB of disk and 1.5TB/sec – 28 PB of disk, >700 GB/sec I/O bandwidth in Lustre bandwith - 12 -

ESnet is a Unique Instrument for Science Unique capabilities ESnet designed for large data - First 100G continental scale network – Connects 40 DOE sites to 140 other - ANI dark fiber can be leveraged to networks develop and deliver 1 terabit – Growing twice as fast as commercial - Services based on user requirements: networks Bandwidth reservations, performance – 50% of traffic is from “big data” monitoring, research testbeds 13

What are the questions driving research in computing? Can we continue the growth in computing performance through more efficient architectures or new paradigms? Limits of Chip Technology What mathematical models, algorithms and software are needed for increasingly complex scientific theories and experimental data sets? Interfaces Key at Mesoscale Can we enable new modes of scientific discovery by applying advanced computing and networking to data from science experiments? Large, Noisy CMB Data

What are the questions driving research in computing? Can we continue the growth in computing performance through more efficient architectures or new paradigms? Limits of Chip Technology What mathematical models, algorithms and software are needed for increasingly complex scientific theories and experimental data sets? Interfaces Key at Mesoscale Can we enable new modes of scientific discovery by applying advanced computing and networking to data from science experiments? Large, Noisy CMB Data

Future of computing: Extreme Heterogeneity? Quantum? Quantum Simulation for Materials, Chemistry and Can new logic devices give us Physics beyond-Moore performance? Mu Multi tiple qu quantum In Investig igatin ing al alternat ative device de devices de technologies te at t Be Berk rkeley Siddiqi’s Quantum Circuit Us Use e Sk Skir irmio ion “b “bags” to act t as info forma rmati tion carri rriers rs Ex Experimentally for fo r mu multi ti-va valued logic devi vice im imple lement c che hemic ical s l sim imula latio ions o on 3-qubit t platfo tform rm Inve vesti tigate te energy y effi ficient t superconducti ting archite tectu tures where info forma rmati tion is sto tored in Deve velop mo model fo for r sma mall user r fa facility ty to to explore magneti ma tic fl flux quanta ta and tr transfe ferr rred with th Single de device technology gy Fl Flux Quantum voltage pulses Fermi Hubbard Quantum Ising, Synthetic gauge Quantum at fractional Bose-Hubbard, fields, Relativistic Chemistry doping Spin-Boson theories THE HAMILTONIAN LANDSCAPE FOR QUANTUM SIMULATION

What are the questions driving research in computing? Can we continue the growth in computing performance through more efficient architectures or new paradigms? Limits of Chip Technology What mathematical models, algorithms and software are needed for increasingly complex scientific theories and experimental data sets? Interfaces Key at Mesoscale Can we enable new modes of scientific discovery by applying advanced computing and networking to data from science experiments? Large, Noisy CMB Data

Transforming how we compute: “smart” math, numerics, HPC: gives unprecedented capability • Smart math leads to science not possible before – Use mathematical properties to build better simulation models – Simulation at previously inaccessible scales – Exploit matrix structure for faster linear algebra • Advanced numerical methodology for better results Numerical simulation emissions in a low swirl burner fueled by hydrogen – High-order discretizations – Projection methodology – Adaptive mesh refinement– resolution where its needed – Surrogate optimization methods • AMReX framework enables HPC solutions – Over 100x increase in throughput MAESTRO simulation near ignition showing flow from center of star and region of high energy generation

What are the questions driving research in computing? Can we continue the growth in computing performance through more efficient architectures or new paradigms? Limits of Chip Technology What mathematical models, algorithms and software are needed for increasingly complex scientific theories and experimental data sets? Interfaces Key at Mesoscale Can we enable new modes of scientific discovery by applying advanced computing and networking to data from science experiments? Large, Noisy CMB Data

The Advanced Light Source (ALS) hosts dozens of different experiments and end station detectors

Data-driven scientific discovery requires integration of modeling, simulation, analysis, data management • Example: 21 st Century Cosmology: – Tight collaboration between astrophysicists and computational scientists to develop new Large Scale Structure simulations technologies for cosmological data of cluster formation in the early universe analysis – Analysis & simulation of 100s of TeraBytes of data from ground- and space-based observations Hydro simulation of a flame front – Modeling & simulation of in a thermonuclear supernova explosion supernovae and large-scale structure formation Palomar Transient Factory data-analysis sky-coverage Cosmic Microwave Background map for the first 3 years of the project Radiation data from Planck

Automated detection and analysis of particle beams in laser-plasma accelerator simulations 375 Machine Learning enables new scientific discoveries from massive data sets Clustering genes and Identifying finding networks hurricanes Image analysis in cosmology and Identifying particle light sources beams in laser plasma simulations Detecting neutrinos Particle simulations: replace with Generative Adversarial Networks Decision support for Modeling human Brain 3D model energy infrastructure behavior reconstruction Projects that advance the state-of-art in machine learning with ties to science •

Welcome to Berkeley Lab Computing Sciences!