AMD introduces open-source HPC Assistant at SC26
AMD introduced an open-source assistant for supercomputing workflows at SC26. The company says it supports job submission, GPU profiling, energy visibility and kernel optimization.
AMD introduced the open-source AMD HPC Assistant at SC26, giving researchers a natural-language interface for system-specific supercomputing guidance. The company says the software can assist with job submission, GPU profiling, energy visibility and kernel optimization.
The first version was developed with CSC, which operates Finland’s LUMI supercomputer, and connected to LUMI’s knowledge base. AMD said it has begun working with Oak Ridge National Laboratory to bring the assistant to Frontier. The company did not give a completion date or describe the scope of the planned deployment.
In AMD’s technical description of the assistant, the company says the system grounds its responses in an HPC center’s documentation, including storage policies and scheduler requirements. For job workflows, AMD says it can synchronize files, generate and submit batch scripts, monitor jobs and retrieve results.
AMD says the profiling workflow exposes compute utilization, cache efficiency and memory traffic to help users identify bottlenecks. Its energy feature uses metrics measured with Omnistat and presents job-level consumption through the assistant and Grafana. For matrix-intensive code, AMD says an optional GEAK workflow can submit and monitor kernel-optimization jobs and surface the results.
The public HPC Assistant repository lists LUMI as the supported site and includes skills for Slurm job management, resource management, AMD GPU profiling and energy-consumption profiling. The repository carries an MIT license and describes GEAK optimization as an optional skill.
More news

AMD-powered Lux supercomputer opens to Oak Ridge researchers

AMD reports $8.5 billion in data-center revenue, forecasts $17 billion for fourth quarter

AMD says Helios production is ramping with initial systems running
