The “Ask the Experts Session #8: AMD Skills” is an in-depth presentation led by Daniel, a presales engineer at AMD, focusing on the AMD Skills ecosystem. The session introduces AMD Skills as a unified platform designed to enhance AI agents’ capabilities in utilizing AMD technology across various hardware, including server and client devices. Daniel emphasizes the goal of making AI agents significantly better at assisting users by providing access to the latest and greatest skills within a cohesive environment. The platform supports a wide range of AMD hardware, starting with Instinct GPUs and expanding to other products, aiming to simplify access and integration through GitHub repositories.
A key highlight of the session is the explanation of the AMD Skills catalog, which includes server-native skills running on Instinct GPUs, collaborative skills spanning client to cloud, and cross-technology skills. Daniel showcases specific examples such as Serving LS on Instinct, which allows deployment without manual intervention, and Serviano Epic, which serves CPUs and other AMD hardware. The skills are designed to optimize performance autonomously, squeezing maximum efficiency out of AMD GPUs, with reported improvements of 30-40% in model performance after a few hours of optimization using AI agents.
The session also covers practical demonstrations of the AMD Skills in action, including the use of TraceLens, a tool developed to profile and analyze neural network models running on AMD hardware. TraceLens identifies bottlenecks in GPU kernels and communication, providing detailed reports that help developers optimize their models. Daniel explains how TraceLens supports various AMD GPUs, including the latest MI300 series, and how it integrates with popular AI frameworks like Codex and Claude. This tool exemplifies AMD’s commitment to transparency and performance tuning, allowing users to understand and improve their AI workloads effectively.
Daniel further discusses the testing and validation processes for AMD Skills, emphasizing rigorous checks to ensure that skills behave as intended without unexpected actions. The team pays special attention to the interaction between different skills within the same workspace to prevent conflicts and ensure smooth operation. This careful curation and continuous testing aim to maintain high-quality assistance for users, whether they are beginners or advanced AI practitioners, ensuring that the skills provide meaningful and reliable support across various use cases.
In conclusion, the session highlights AMD’s strategic approach to building a robust AI skills ecosystem that leverages their hardware capabilities to deliver enhanced AI agent performance. Daniel invites participants to explore the skills catalog, try out the skills via GitHub or integrated platforms, and engage with the community for feedback and collaboration. The presentation underscores AMD’s dedication to innovation, user support, and continuous improvement, with upcoming tools and expansions planned to further empower AI developers and users within the AMD ecosystem.
Useful Links
