Feature Overview
We need to provide a software, which is capable of partitioning vGPUs into different MIG configurations in a way that is consumable for Compute services such as Nova (near future) and Cyborg (far future). The feature will also provide a documentation on how to preconfigure MIG before using vGPU MIG-backed instances in RHOSO. Proper preconfiguration of MIG ensures that users can fully leverage MIG-backed vGPUs for their workloads with optimized resource partitioning and performance.

Goals

• To enable Compute service(s) (such as Nova and Cyborg) to manage MIG configuration and apply such configuration easily without additional manual pre-steps.
• To provide documentation for users how to use this software within RHOSO product

Who benefits from this Feature, and how?

• Customers coming from other virtualization platforms will have c-series MIG-backed parity
• AI/ML users benefit from clear guidance on setting up MIG for optimal GPU utilization.
• RHOSO administrators can correctly configure MIG to avoid misconfigurations and ensure predictable GPU resource allocation.
• Organizations can improve GPU efficiency by following best practices for MIG setup.
• Data Scientist can run more inference services with more MIG slices with memory isolation.

What is the difference between today’s current state and a world with this Feature?

• Current State: No official RHOSO documentation exists for preconfiguring MIG before enabling vGPU MIG-backed instances.
• Future State: Users have clear and detailed guidance on how to configure MIG properly before deploying workloads.

Requirements :

Done - Acceptance Criteria 

• Suitable SW/driver is part of RHOSO product and appropriate installation steps are part of Ansible module(s) of RHOSO installers
• User documentation of MIG management software is part of downstream documentation

Use Cases - i.e. User Experience & Workflow:

• A system administrator wants to enable MIG on NVIDIA GPUs before deploying vGPU MIG-backed workloads in RHOSO.
• A user follows the documentation to partition a GPU using MIG and validate the setup. A troubleshooting guide helps users resolve issues when MIG is not detected correctly.

Out of Scope:

• Installation logic on OCP operator level as this will be added as part of deployment of specific service which will be use the MIG management software/driver
• Implementation of MIG-backed vGPU support itself.
• Changes to RHOSO scheduling mechanisms for vGPU resources.

Documentation Considerations

• Create a step-by-step guide with screenshots or CLI examples.
• Provide verification commands and expected output for confirming MIG setup.
• Link to NVIDIA official documentation where relevant.

Questions to Answer:

• What are the supported GPU models for MIG preconfiguration in RHOSO?
• Are there any dependencies or prerequisites users need to be aware of before enabling MIG?
• What are the recommended best practices for configuring MIG for AI/ML workloads?

Background and Strategic Fit:

• Ensuring MIG is properly preconfigured is a prerequisite for enabling vGPU MIG-backed support in RHOSO.
• Providing clear documentation prevents user misconfigurations and improves overall GPU resource efficiency.
• Aligns with industry best practices for AI/ML and HPC workload management.

Customer Considerations:

• Customers who rely on NVIDIA MIG for workload partitioning need official guidance on setup.
• Organizations deploying RHOSO in multi-GPU environments benefit from clear instructions to optimize resource allocation.

Team Sign Off (Completion while in Planning status)

• All required Epics (known at the time) are linked to the this Feature
• All required Stories, Tasks (known at the time) for the most immediate Epics have been created and estimated
• Add - Reviewers name, Team Name
• Acceptance == Feature as “Ready” - well understood and scope is clear - Acceptance Criteria (scope) is elaborated, well defined, and understood
• Note: Only set FixVersion/s: on a Feature if the delivery team agrees they have the capacity and have committed that capability for that milestone

• …