Reference Stacks#
AMD Enterprise AI Reference Stack#
The AMD enterprise AI reference stack is an opinionated software stack for developing, deploying, and running AI workloads on a Kubernetes platform. The platform can be utilized by system administrators, resource managers, AI researchers, and AI solution developers.
The AMD enterprise AI reference stack brings together:
AMD Inference Microservices (AIMs): Portable inference microservices for serving models on AMD GPUs. AIMs provide a mechanism for automatic selection of optimized runtime parameters based on user input, hardware, and model specifications.
AIM Engine: A Kubernetes operator for deploying and managing inference workloads.
AMD AI Workbench: An easy-to-use graphical interface for running and managing AI workloads, including features such as one-click deployment of AIMs from the AIM Catalog and fine-tuning capabilities.
AMD Resource Manager: Provides administrators with tools to manage clusters, monitor resources, and maintain team access to computational resources.
Together, these components enable capabilities such as:
High GPU efficiency: Intelligent workload placement and dynamic resource sharing
Unified AI infrastructure: Standardized governance, tools, and processes enabling seamless collaboration across teams
Accelerated time to value: Built-in inference and streamlined development workflows
AI-native orchestration: Intelligent job prioritization, dynamic allocation, and consistent performance
Information security: Built-in RBAC with the possibility to federate and integrate with existing SSO and IAM solutions.
The reference stack supports bare-metal, on-premises, and public cloud deployments of AI workloads. It has been validated on Instinct MI300X, MI325X, MI350X and MI355X GPUs. To get started, view the resources below.
Resource |
Link |
|---|---|
GitHub |
|
Getting started (on-premises) |
|
Getting started (DigitalOcean) |
|
Getting started (Workbench) |
|
Getting started (Resource Manager) |