In this session, AMD experts introduced the AMD AI Playbooks, a curated collection of continuously tested, step-by-step guides that help developers run AI workloads efficiently on various AMD hardware platforms, including the Ryzen AI HaloBox. They demonstrated the Playbooks’ dynamic adaptability, showcased a live AI image generation demo, and highlighted community contributions and the AMD AI developer program supporting collaboration and innovation.
In this Ask the Experts session, Ben Consalvo and Sreeram from AMD introduce the AMD AI Playbooks, a collection of step-by-step guided walkthroughs designed to help developers run real AI workloads on AMD hardware. These playbooks are not just manuals or documentation but curated paths that combine environment setup, scripts, commands, and educational content about AI models and techniques. The playbooks cover a variety of popular open-source frameworks and are intended to kickstart developers’ journeys on AMD devices, supporting a range of hardware including Ryzen AI HaloBox, Ryzen AI APUs, and Radeon GPUs across Windows and Linux operating systems.
Sreeram showcases the AMD Ryzen AI HaloBox, a compact developer platform that enables running large AI models locally without cloud dependency, ensuring data privacy and eliminating usage costs. The HaloBox comes pre-installed with necessary software and models, providing a zero-setup experience. He also demonstrates the Playbooks website, highlighting its filtering capabilities by hardware, OS, and difficulty level, and explains how the instructions adapt dynamically based on the selected device and operating system. This flexibility makes it easy for users to find and follow relevant playbooks tailored to their specific hardware.
A key feature emphasized is the rigorous continuous integration (CI) testing process that ensures all playbooks remain up-to-date and functional. Unlike many tutorials that become outdated quickly, AMD’s playbooks are automatically tested nightly on real AMD hardware with the latest software versions. The actual commands shown on the website are the same ones executed during testing, guaranteeing reliability and reducing user frustration. This approach supports multiple operating systems and hardware targets from a single source, maintaining high quality and accuracy across the board.
Ben then provides a detailed live demo of the ComfyUI playbook, which uses the Z Image Turbo diffusion model to generate images locally on AMD hardware. He explains the components of the model pipeline, including the text encoder, diffusion model, and variational autoencoder, and demonstrates how the model produces high-quality images with legible text and realistic reflections. He also discusses key parameters like the number of diffusion steps and guidance scale, showing their impact on image quality and generation speed. The demo highlights the advantages of running AI models locally without cloud costs or data privacy concerns.
Finally, the session covers how the community can contribute new playbooks through a streamlined proposal and review process to maintain quality standards. Ben mentions ongoing efforts to translate playbooks into multiple languages and encourages feedback on desired new playbooks. The presentation concludes with information about the AMD AI developer program, which offers cloud credits, sweepstakes, and direct access to AMD engineers, fostering a collaborative environment for AI developers using AMD hardware. The session ends with a Q&A addressing platform support and other user inquiries.