Dell, Nvidia, AMD to power new Ohio AI compute cluster

Aims to develop next generation cyberinfrastructure, amongst other things

The Ohio Supercomputer Center (OSC) is building out a new high performance compute cluster for AI applications based on Dell hardware with AMD Epyc processors and Nvidia GPU accelerators.

Known as Ascend, the new HPC cluster is set to launch later this year to support artificial intelligence, machine learning, big data and data analytics work at the OSC, which is known for public-private partnerships and industrial HPC. It joins the site's existing Pitzer and Owens compute clusters, and will triple the facility's capacity for AI, modeling and simulation, according to the Center.

OSC has experienced growing demand for GPU resources over the past several years, according to the site's associate director Doug Johnson, and so Ascend will be its first computing cluster devoted entirely to providing intensive GPU-based processing capabilities.

"By establishing a cluster focused on analysis of very large datasets quickly, support for classes of AI/ML applications that can't run on our current systems, and simulations that require the fastest GPUs, OSC will better meet the needs of these clients while ensuring the prompt processing of requests for our existing clusters, Owens and Pitzer," he said.

Hardware for the new cluster comprises 24 Dell PowerEdge XE8545 server nodes, 4U rackmount systems with dual AMD Eypc 3rd-Gen processors. Each node will be configured with four Nvidia A100 80GB GPUs, and interconnected with Nvidia Quantum HDR 200Gbps networking.

The site is currently involved in two US National Science Foundation-funded projects designed to advance AI work. The first is the AI Institute for Intelligent Cyberinfrastructure with Computational Learning in the Environment (ICICLE), which is led by Ohio State University and aims to develop next generation cyberinfrastructure with a focus on making AI more accessible.

The second is an "AI Bootcamp for Cyberinfrastructure (CI) Professionals" to be held this year, which will focus on building skills in AI among staff at research computing facilities like OSC nationally.

Karen Tomko, the OSC's director of research software applications, said that Ascend will provide a state-of-the art resource for the ICICLE research team to explore and develop new AI technologies while also giving the OSC's own staff an opportunity to increase their understanding of the AI workload and best practices for support of this growing area.

Ascend will be accessible through OSC's OnDemand web portal, but the site's client services team will evaluate requests to run projects on the cluster. The OSC said it will reveal a more detailed timeline for the Ascend launch and availability this summer. ®