Quantum computers must operate in concert with classical computers to deliver on the promise of quantum advantage for practical problems. To achieve that, it is important to understand how quantum and classical computing can interact together, and how one can characterize the scalability and efficiency of hybrid quantum–classical workflows. So far, early experiments with quantum-centric supercomputing workflows have been limited in scale and complexity. Here, we use a Heron quantum processor deployed on premises with the entire supercomputer Fugaku to perform the largest computation of electronic structure involving quantum and classical high-performance computing. We design a closed-loop workflow between the quantum processors and 152,064 classical nodes of Fugaku, to approximate the electronic structure of chemistry models beyond the reach of exact diagonalization, with accuracy comparable to some all-classical approximation methods. Our work pushes the limits of the integration of quantum and classical high-performance computing, showcasing computational resource orchestration at the largest scale possible for current classical supercomputers.

Vela is a cloud-native system designed for LLM training workloads built using off-the-shelf hardware, Linux KVM-based virtualization, and a virtualized RDMA over Converged Ethernet (RoCE) network. Vela virtual machines (VMs) support peer-to-peer DMA between the GPUs and SRIOV-based network interface. In this paper, we share Vela's key architectural aspects with details from an NVIDIA A100 GPU-based deployment in one of the IBM Cloud data centers. Throughout the paper, we share insights and experiences from designing, building, and operating the system over a ~2.5 year timeframe to highlight the capabilities of readily available software and hardware technologies and the improvement opportunities for future AI systems, thereby making AI infrastructure more accessible to a broader community. As we evaluated the system for performance at ~1500 GPU scale, we achieved ~80% of the ideal throughput while training a 50 billion parameter decoder model using model parallelism, and ~70% per GPU FLOPS compared to a single VM with the High-Performance Linpack benchmark.

With Moore's law progressively running out of steam, heterogeneous computing architectures have been powering the top supercomputers in the world for many years and are now finding broader adoption across the industry. The trend towards sustainable computing also requires domain-specific heterogeneous hardware architectures, which promise further gains in energy efficiency. At the same time, today's high performance computing applications have evolved from monolithic simulations in a single domain to multidisciplinary complex workflows. In this paper, we explore how these trends affect system design decisions and what this means for future computing system architectures.

Uncertainty prediction for ocean and climate predictions is essential for multiple applications today. Many-Task Computing can play a significant role in making such predictions feasible. In this manuscript, we focus on ocean uncertainty prediction using the Error Subspace Statistical Estimation (ESSE) approach. In ESSE, uncertainties are represented by an error subspace of variable size. To predict these uncertainties, we perturb an initial state based on the initial error subspace and integrate the corresponding ensemble of initial conditions forward in time, including stochastic forcing during each simulation. The dominant error covariance (generated via SVD of the ensemble) is used for data assimilation. The resulting ocean fields are used as inputs for predictions of underwater sound propagation. ESSE is a classic case of Many Task Computing: It uses dynamic heterogeneous workflows and ESSE ensembles are data intensive applications. We first study the execution characteristics of a distributed ESSE workflow on a medium size dedicated cluster, examine in more detail the I/O patterns exhibited and throughputs achieved by its components as well as the overall ensemble performance seen in practice. We then study the performance/usability challenges of employing Amazon EC2 and the Teragrid to augment our ESSE ensembles and provide better solutions faster.

This work presents a novel method to detect an important flow feature, vortices, in the ocean. Our method can detect closed streamlines around vortex cores. Coupled with existing vortex core detection, the entire vortex area, which is the combination of the vortex core and surrounding streamlines, can be detected. A variety of feature extraction methods are presented, and those more pertinent to this study are implemented. Detection results are evaluated in terms of accuracy, clarity and usability.

Direct Numerical Simulation (DNS) of turbulence requires many CPU days and Gigabytes of memory.These requirements limit most DNS to using supercomputers, available at supercomputer centres.With the rapid development and low cost of PCs, PC clusters are evaluated as a viable low-cost option for scientific computing.Both low-end and high-end PC clusters, ranging from 2 to 128 processors, are compared to a range of existing supercomputers, such as the IBM SP nodes, Silicon Graphics Origin 2000, Fujitsu AP3000 and Cray T3E.The comparison concentrates on CPU and communication performance.At the kernel level, BLAS libraries are used for CPU performance evaluation.Regarding communication, the free implementations of MPICH and LAM are used on fast-ethernet-based systems and compared to myrinet-based and supercomputer networks.At the application level, serial and parallel simulations are performed on state of the art DNS, such as turbulent wake flows in stationary and moving computational domains.

In this paper we analyze communication patterns in the parallel three-dimensional Navier-Stokes solver Prism, and present performance results on the IBM SP2, the Cray T3D and the SGI Power Challenge XL. Prism is used for direct numerical simulation of turbulence in non-separable and multiply-connected domains. The numerical method used in the solver is based on mixed spectral element-Fourier expansions in (x-y) planes and z-direction, respectively. Each (or a group) of Fourier modes is computed on a separate processor as the linear contributions (Helmholtz solves) are completely uncoupled in the incompressible Navier-Stokes equations; coupling is obtained via the nonlinear contributions (convective terms). The transfer of data between physical and Fourier space requires a series of complete exchange operations, which dominate the communication cost for small number of processors. As the number of processors increases, global reduction and gather operations become important while complete exchange becomes more latency dominated. Predictive models for these communication operations are proposed and tested against measurements. A relatively large variation in communication timings per iteration is observed in simulations and quantified in terms of specific operations. A number of improvements are proposed that could significantly reduce the communications overhead with increasing numbers of processors, and {\em generic} predictive maps are developed for the complete exchange operation, which remains the fundamental communication in Prism. Results presented in this paper are representative of a wider class of parallel spectral and finite element codes for computational mechanics which require similar communication operations.