AMD Brings Agentic AI to Supercomputing with Open-Source AMD HPC Assistant, Debuting on LUMI and Frontier

Nov 16, 2026

HPC Assistant

Introducing AMD HPC Assistant, an open-source, AI-powered guide that helps scientists move faster from code to discovery

AMD powers the world's leading scientific computing, from the exascale supercomputers driving discovery in the United States and Europe to Lux, the first Genesis Mission system for AI-driven science. With leadership CPUs, GPUs, FPGAs, networking and open ROCm software, AMD gives nations, laboratories and enterprises a single open platform for high-performance computing, AI for science, sovereign AI, agentic AI and the path to quantum computing. AMD HPC Assistant extends that innovation to the researcher experience, applying agentic AI to help scientists use supercomputers more productively and efficiently.

Supercomputers such as LUMI and Frontier enable climate modeling, molecular simulation, genomic analysis, AI, and other workloads at an extraordinary scale. Yet powerful infrastructure does not automatically deliver scientific productivity. Researchers can face steep learning curves, system-specific policies, complex schedulers, software configuration, debugging, performance bottlenecks, and limited energy visibility. HPC support teams spend valuable time addressing many of these same challenges.

AMD HPC Assistant can help close that gap, bringing system-aware HPC expertise into the AI coding tools researchers already use, accelerating the path from scientific intent to an optimized workload.

What Is AMD HPC Assistant?

AMD HPC Assistant is an open-source AI assistant that brings HPC knowledge and workflow automation into a natural-language experience. Instead of navigating documentation, support portals, terminals, profiling tools, and dashboards, researchers can describe what they need and receive guidance grounded in their actual HPC system.

Think of it as an experienced HPC engineer inside the workflow. The assistant understands system directories, storage quotas, purge policies, scheduler requirements, and GPU profiling practices, translating user objectives into commands and job scripts.

Accurate, system-specific guidance depends on connecting the assistant to the HPC center's own knowledge base, so responses reflect the actual environment researchers are working in.

The first version was developed with CSC, the Finnish IT Center for Science, which operates LUMI. AMD HPC Assistant has been connected to LUMI's knowledge base, enabling it to answer questions grounded in LUMI's own documentation, storage policies, scheduler requirements, and site-specific guidance. The collaboration targets practical challenges including onboarding, job submission, site-specific requirements, and performance diagnosis without requiring every scientist to become an HPC systems expert.

Building on LUMI, AMD has begun working with Oak Ridge National Laboratory to bring HPC Assistant to Frontier, the world’s first exascale supercomputer.

From Scientific Question to Optimized Result

AMD HPC Assistant goes beyond conversational access to documentation. It creates an agentic workflow across scientific computing, helping researchers move code and data to the system, prepare environments, submit and monitor jobs, analyze GPU performance and energy, and optimize kernels.

Get answers grounded in the actual system

Every HPC center has different filesystems, quotas, schedulers, policies, and software. AMD HPC Assistant searches the center’s approved documentation for site-specific answers. Ask where to place a 100 GB dataset, for example, and it can provide the appropriate directory, quota, purge policy, and storage guidance, reducing user guesswork and repetitive support requests.

Transfer files and manage jobs through natural language

Researchers describe what they want to run, while the assistant helps synchronize files, generate and submit batch scripts, monitor jobs, and retrieve results. This removes avoidable friction between scientists and the infrastructure, shortening the path from local code to completed HPC workloads.

Understand GPU performance in depth

GPU profiling gives researchers a detailed view of how their workloads use the hardware, including compute utilization, cache efficiency, and memory traffic. This helps them pinpoint what is limiting performance and get actionable guidance on where to optimize.

Make energy consumption visible

Energy Profiling connects to the HPC system metrics measured with Omnistat1 and presents job-level consumption through the assistant and Grafana2. Researchers can compare runs, refine workloads, and make better-informed use of compute and energy resources.

Optimize kernels for AMD hardware

GEAK1 kernel optimization helps tune matrix-intensive kernels for AMD GPU architectures. The assistant submits and monitors optimization jobs and surfaces results, bringing advanced optimization closer to scientists without requiring them to manually manage every low-level step.

Launch purpose-built scientific AI agents (preview at SC26)

AMD HPC Assistant can launch domain-specific AI agents and AMD Inference Microservices (AIMs) for scientific workloads. On LUMI, for example, the OpenFold3 AIM enables protein-structure prediction. Researchers provide an amino-acid sequence, while the assistant starts the GPU server, downloads model weights, submits the job, and guides visualization without manual environment configuration.

Containerize and deploy AI models with AIM Agent (preview at SC26)

AIM Agent helps teams bring their own AI models to HPC. It reads model repositories and documentation, identifies requirements, creates build recipes and container images, tests and benchmarks them, and generates deployment-ready reports, helping to turn days of manual configuration into a structured, repeatable workflow.

Open Source by Design for Shared Innovation

Open source is not an afterthought at AMD. It is a core architectural choice.

Scientific computing advances through collaboration, transparency, portability, and community innovation. HPC centers should be able to shape their AI-assisted environments around their own infrastructure, security policies, schedulers, storage, documentation, models, and users.

With an open-source foundation, centers can inspect and adapt the platform, contribute system-specific capabilities, and integrate new agents and models as requirements evolve. Research institutions, technology partners, and scientific communities can contribute to innovation while each site retains control over what it deploys and operates.

HPC Assistant is open source and available on GitHub, where you can explore the code, configuration options, and documentation to get started in your own environment. If you would like to see it in action, stop by the AMD booth #3028 at SC26 in Chicago, where we will be running demos throughout the conference.

Help Shape the Next HPC Experience

If you operate an HPC center, build scientific applications, develop domain AI models, or want to contribute to an open HPC assistant ecosystem, we want to work with you.

Join the effort. Bring AMD HPC Assistant to your environment, contribute to the open-source platform, and help define how scientists engage with the next generation of supercomputers.

Contact HPCAI@AMD.com to start the conversation.

Footnotes

1. GEAK (Generating Efficient AI-Centric Kernels) is an autonomous, multi-agent optimization framework developed by AMD to accelerate GPU code performance across AMD Instinct (MI300X, MI355X, MI250), RDNA3.5, and RDNA4 architectures.

Share:

Article By


Contributors


Sr. Product Marketing Manager

Related Blogs