# AMD Launches Open HPC Assistant on LUMI as Frontier Work Begins

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/59080
Published: 2026-11-17T04:40:11.000Z
Updated: 2026-11-17T04:40:11.000Z
Section: Technology

> The software supports documentation, job management and GPU analysis. LUMI is its only supported site, and no Frontier rollout date has been announced.

Advanced Micro Devices (AMD) announced an open-source assistant for high-performance computing (HPC) researchers Nov. 16, with LUMI as its only supported system and work underway to adapt the software for Frontier, the U.S. supercomputer.

The first version was developed with CSC, the Finnish IT Center for Science, which operates LUMI. It connects a coding agent to LUMI’s system documentation, allowing researchers to ask questions about procedures specific to that machine. The setup is designed to provide operational guidance tied to local instructions and resources.

The assistant’s listed tasks span several stages of a computing job. It can help synchronize files, prepare and submit job scripts, monitor workloads, and examine GPU performance through profiling. Users can also view energy use for an individual job. Documentation about storage, quotas and scheduling gives the assistant context for system-specific requests.

The product is software for working with a supercomputer; it does not add computing capacity. Its guidance for a particular site depends on access to that center’s documentation. Researchers use the underlying system to run their workloads, while the assistant helps them handle tasks around those runs.

LUMI is currently the assistant’s only supported site. AMD and Oak Ridge National Laboratory (ORNL) have begun adapting it for Frontier, but no rollout date has been announced. Frontier is the United States’ first exascale system.

“We are excited to work with AMD to bring HPC Assistant to Frontier,” said Dr. Antigoni Georgiadou, a computational scientist at ORNL.

AMD has made the code public. AMD-authored software is licensed under MIT, with third-party components identified separately. Setup requires an AI backend; available configurations include an Aitta token for LUMI or an Anthropic API key with outbound network access. The software also supports OpenAI-compatible endpoints.


