To prove quantum supremacy, a joint research team from Google Inc., NASA Ames Research Center, and the Department of Energy’s Oak Ridge National Laboratory (ORNL), needed to rule out that classical supercomputers could perform computational tasks at the same speed as Google’s Sycamore quantum computer. So, they came to the IBM AC922 Summit, the world’s fastest and smartest classical supercomputer for open science. Summit is managed by the Oak Ridge Leadership Computing Facility (OLCF), a US Department of Energy (DOE) Office of Science User Facility located at ORNL.
“To test quantum supremacy, we needed a well-defined computational problem, a benchmark task that you could run on both quantum and classical computers, to then observe which solved it faster,” said Dmitry Lyakh, a computational scientist at the OLCF.
A software library developed by ORNL’s Dmitry Lyakh allowed the team to take full advantage of the Summit supercomputer to run a quantum benchmark code. Credit: Genevieve Martin/ Oak Ridge National Laboratory, U.S. Dept. of Energy
The quantum supremacy test, which was run on a 53 qubit quantum computer, sampled a random quantum circuit with a depth of 20. To prove quantum supremacy, however, the team needed to simulate that same quantum circuit on Summit, a heterogeneous GPU-accelerated supercomputer.
But there was a hurdle. This test, which scientists call random circuit sampling (RCS), was written in a code optimized to run on regular CPU-only computer clusters.
That’s when Lyakh sprang into action, adapting the classical RCS code—originally developed by Google and NASA and dubbed qFlex—to be executed on Summit’s multi-GPU heterogeneous nodes.
RCS calculations require that computers perform multiplications of tensors, which are generalizations of matrix multiplication in linear algebra. Adapting the code for Summit required offloading all numerical tensor algebra computations programmed in qFlex to Summit’s GPUs, which was done using a software library developed by Lyakh called TAL-SH.
The library was developed by the OLCF at the Summit Center for Accelerated Application Readiness, sponsored by DOE’s Advanced Scientific Computing Research program, and was later extended for use as a computational back end for the qFlex code.
This highly optimized library unleashed enormous computing power, delivered by Summit’s NVIDIA Volta GPUs, to perform the computationally intensive tensor operations required for a classical simulation of the RCS circuits with the qFlex code.
Thanks in part to Lyakh’s contribution, the team was able to determine that the computation that took Google’s Sycamore quantum computer only 200 seconds would have taken Summit 10,000 years to complete with current state-of-the-art algorithms.
This historic result will open the door to answers to a new array of questions for computer scientists.
“The next milestone is to build a quantum computer that is large enough to begin to solve problems of practical interest, everyday-life problems, the same way Summit currently does, but even faster,” said Lyakh. “That’s where we are going next.”
Related Publication: F. Arute et al., “Quantum Supremacy Using a Programmable Superconducting Processor.” Nature574 (2019), doi:10.1038/s41586-019-1666-5.
The OLCF, home of Summit, is a US Department of Energy (DOE) Office of Science User Facility at DOE’s Oak Ridge National Laboratory (ORNL). UT-Battelle LLC manages Oak Ridge National Laboratory for DOE’s Office of Science, the single largest supporter of basic research in the physical sciences in the United States. DOE’s Office of Science is working to address some of the most pressing challenges of our time. For more information, visit https://energy.gov/science.
