Running Popular HPC and AI Applications on Fuzzball

July 5, 2023

Managing complex simulations can be overwhelming for users who lack a deep understanding of HPC solutions. These users may struggle with setting up the simulations, configuring parameters, managing resources, and interpreting the results. The growing complexity of simulations makes it essential to find ways to simplify the process. Fortunately, we have a solution: Fuzzball, CIQ’s next-generation High Performance Computing (HPC) platform, simplifies how simulation users interact with their compute resources.

Fuzzball unites the rapidly converging use cases in performance-intensive computing by providing a single platform that deftly combines the best of cloud, hyperscale, and enterprise with that of traditional HPC, creating an entirely new computing paradigm that we call “HPC 2.0.” The software offers a turnkey, Kubernetes-based, hybrid computing infrastructure stack that empowers users with more control, administrators with more reach, and organizations with more productivity.

The beauty behind this solution is that Fuzzball makes it possible for users to knit together workflows and define their data movement. It automates application execution, data movement, and all the low-level specifics so that scientists and researchers can focus on their work.

To show how users can deploy Fuzzball within their organizations, CIQ’s engineers successfully tested a number of popular open source applications and workflows:

OpenFOAM: ran an end-to-end pipeline of the simulation of the well-known motorbike sample workload and visualization of the results using ParaView
OpenRadioss: ran full end-to-end simulations of numerous workloads provided by the OpenRadioss community and visualization of the results using ParaView
TensorFlow and PyTorch: built and executed workloads using these machine learning frameworks, demonstrating the ability of Fuzzball to combine workloads across the HPC/AI space
Quantum ESPRESSO, GROMACS, LAMMPS: tested popular and widely used codes in research computing