At the hybrid virtual/in-person SC21 conference held recently, from Zhijiang Lab,The 14-member Chinese research team of Tsinghua and Wuxi Supercomputing Center won the Gordon Bell Prize, the highest award in the supercomputing field this year, for simulating quantum circuits with the new exascale Sunway system.
It is understood that the GORDON BELL PRIZE, established in 1987, is the highest academic award in the field of high-performance computing applications in the world. It is called “the Nobel Prize in the field of supercomputing” and is mainly awarded to high-performance applications. The most outstanding achievement in the field, selected and awarded by ACM every year, has a large international influence. The winner will receive a prize of US$10,000 from HPC pioneer Gordon Bell.
ACM aims to “track the progress of parallel computing over time, with special emphasis on the application of high-performance computing to science, engineering, and large-scale data analysis innovation rewards.”. This year’s Gordon Bell Prize finals have 6 projects, involving quantum computing, molecular dynamics, spectroscopy and nuclear fusion and other fields.
Attachment: 2021 ACM Gordon Bell Prize winner
Bridging the “quantum hegemony” gap: using the new Sunway supercomputer to realize random quantum circuit simulation
14 researchers from Zhejiang Laboratory, Tsinghua University, Wuxi National Supercomputing Center and Shanghai Quantum Science Research Center used the large-scale new Sunway Exascale System to conduct groundbreaking circuit simulations of quantum.
“With Google’s declaration of “Quantum Hegemony” in 2019, claiming that the Sycamore superconducting quantum computer is more than one billion times faster than Summit, the dawn of the quantum era begins to unfold in a more positive way,” the researchers wrote, “IBM Research The team later responded that they could complete the simulation on the classic Summit supercomputer… within a few days, not 10,000 years.”
The team used quantum state sampling of random quantum circuits as an example to solve this controversial quantum advantage. The researchers’ random quantum circuit simulator, combined with the powerful functions of the Sunway exascale system, simulated a 10×10 (qubit)×(1+40+1) (depth) circuit, with a sustained performance of 12 billion billion times. For single-precision calculations or 44 billion billion mixed-precision calculations, researchers call this “a new milestone in the classical simulation of quantum circuits.” They reduced the analog sampling time from the previously estimated 10,000 years to 304 seconds.
*The above picture is provided by the researchers, and the main classical random quantum circuit simulation summary.The X axis represents the number of qubits; the Y axis represents the corresponding memory space required; the size of the circle and rectangle represents the complexity/depth of the circuit.
Nominations for the ACM Gordon Bell Award:
While quantum simulation research has won awards, the other five nominations represent the most in-depth research on some of the most urgent research applications in the world.
·Anton 3: 20 microsecond molecular dynamics simulation before lunch
As many as 67 researchers participated in this research and eventually developed the dedicated Anton 3 molecular dynamics supercomputer designed and built by DE Shaw. Researchers report that Anton 3 can simulate one million atoms on 512 nodes at a rate of 100 microseconds per day, and that the energy per simulated microsecond is an order of magnitude less than any other machine. To achieve this feat, they implemented a series of architecture and algorithm improvements, including a new custom network, specialized pairwise interactions with different precisions, and a “Manhattan method” for solving unbound interactions. new method.
·The 400 trillion grid Vlasov simulation on the “Fugaku” supercomputer: the large-scale distribution of neutrinos in the six-dimensional phase space in the remains of the universe
The researchers used large-scale simulations of neutrinos in cosmic relics, combined with N-body simulations of cold dark matter. Their largest simulation spanned 400 trillion grids and 330 billion celestial bodies, “accurately reproduced the nonlinear dynamics of neutrinos in the universe.” After optimization on “Fuyue”, the researchers expanded 147,456 Nodes, showing up to 96% weak scaling and up to 93% strong scaling.
Full-volume simulation of symplectic structure particles with 111.3 trillion particles and 25.7 billion grids in tokamak plasma
Tokamak, a toroidal container that uses magnetic confinement to achieve controlled nuclear fusion. More than a dozen researchers from China used the new Sunway system to simulate the tokamak’s full-volume confined toroidal plasma. These simulations reached 111.3 trillion particles and 25.7 billion grids, achieving sustained performance of more than 201 petaflops double precision, and the fastest iteration step number reached 298.2.
·Ab Initio quantum Raman spectroscopy simulation of China’s leading high-performance computing system
This research also uses the new Shenwei Exascale System to push Raman spectroscopy-a structural fingerprint-to new limits. The dozens of researchers from China explained that “Raman spectroscopy provides chemical and compositional information, which can be used as a structural fingerprint of various materials. Therefore, the simulation of Raman spectroscopy including quantum perturbation analysis and ground state calculations All are of great significance.” The full quantum mechanical simulation of the Raman spectra of biological materials has proved to be particularly difficult. Here, the researchers conducted “fast, accurate, and massively parallel real biological system Raman spectroscopy all ab initio simulations.” , Up to 3006 atoms, 468.5 petaflops for double precision, and 813.7 petaflops for mixed half-precision, which shows “the new application potential of the QM method in biological systems”.
·Billion-atom molecular dynamics simulation of carbon under extreme conditions and experimental time and length scales
These researchers observed the BC8 phase of carbon “long-term search” under extreme pressure and temperature. To this end, they ran on the U.S. Leadership Summit system, commanded 4650 nodes within 24 hours, and demonstrated the “SNAP [分子动力学] Unprecedented scalability and unparalleled real-world performance.” Represents the simulation of physical time in nanoseconds. The simulation has achieved a parallel efficiency of more than 97% and a peak computing power of 50 petaflops. Molecular dynamics simulation. According to the researchers, this record is 6.2 million atomic steps per node second, which is 22.9 times the previous record.