Application Experiences on a GPU-Accelerated Arm-based HPC Testbed.

This paper assesses and reports the experience of ten teams working to port, validate, and benchmark several High Performance Computing applications on a novel GPU-accelerated Arm testbed system. The testbed consists of eight NVIDIA Arm HPC Developer Kit systems, each one equipped with a server-class Arm CPU from Ampere Computing and two data center GPUs from NVIDIA Corp. The systems are connected together using InfiniBand interconnect. The selected applications and mini-apps are written using several programming languages and use multiple accelerator-based programming models for GPUs such as CUDA, OpenACC, and OpenMP offloading. Working on application porting requires a robust and easy-to-access programming environment, including a variety of compilers and optimized scientific libraries. The goal of this work is to evaluate platform readiness and assess the effort required from developers to deploy well-established scientific workloads on current and future generation Arm-based GPU-accelerated HPC systems. The reported case studies demonstrate that the current level of maturity and diversity of software and tools is already adequate for large-scale production deployments.

Parallel computational steering for HPC applications using HDF5 files in distributed shared memory.

Interfacing a GUI driven visualization/analysis package to an HPC application enables a supercomputer to be used as an interactive instrument. We achieve this by replacing the IO layer in the HDF5 library with a custom driver which transfers data in parallel between simulation and analysis. Our implementation using ParaView as the interface, allows a flexible combination of parallel simulation, concurrent parallel analysis, and GUI client, either on the same or separate machines. Each MPI job may use different core counts or hardware configurations, allowing fine tuning of the amount of resources dedicated to each part of the workload. By making use of a distributed shared memory file, one may read data from the simulation, modify it using ParaView pipelines, write it back, to be reused by the simulation (or vice versa). This allows not only simple parameter changes, but complete remeshing of grids, or operations involving regeneration of field values over the entire domain. To avoid the problem of manually customizing the GUI for each application that is to be steered, we make use of XML templates that describe outputs from the simulation (and inputs back to it) to automatically generate GUI controls for manipulation of the simulation.