Repository navigation
Wiki ‐ Hyperscale GPU Management Workstream
Disclaimer: Please do not submit any confidential information to the Project Community. All presentation materials, proposals, meeting minutes and/or supporting documents are published by OCP and are open to the public in accordance to OCP's Bylaws and IP Policy. This can be found on the OCP OCP Policies page. If you have any questions please contact OCP.
Welcome to the OCP Hyperscale GPU Management Workstream.
This workstream is a closed CLA Workstream under OCP Hardware Management Project
This CLA workstream is focused on defining the compliance requirements and specifications for managing hyperscale GPU systems in data centers. The goal is to create a standardized approach that ensures interoperability, efficiency, and scalability across different hardware and software platforms. This workstream will collaborate with other OCP projects and industry partners to develop best practices, reference designs, and open-source tools that facilitate the deployment and management of hyperscale GPU infrastructures.
To simplify the process of contributing to this workstream, we have broken down the work into 3 tracks to independently focus on GPU FW Update, GPU Managebility Interfaces & Telemetry, and GPU RAS. To supplement this CLA Workstream, we also have a dedicated track for building the compliance tool, CTAM with the specification to define compliance scenarios.
We encourage active participation from the community and welcome feedback, suggestions, and contributions from all stakeholders.
The CLA is comprised of 5 organizations as of today: Microsoft, Meta, Google, NVIDIA, and AMD. The following individuals are the current workstream leads:
GPU FW Update Track Lead: Sujoy Sen - Google
GPU Managebility Interfaces & Telemetry Track Lead: Justin York - Google
GPU RAS Track Lead: Rama Bhimanduni - Microsoft
In addition the GPU compliance tool track, CTAM, is led by: Rajat Madhusudan - Microsoft.
OCP GPU FW Update Specification: This document specifies the GPU firmware(FW) update requirements and Redfish/PLDM interfaces to perform FW updates on GPU products. This document covers the FW Update model for discrete GPUs and Universal Base-board based GPU systems containing management controllers. This document further captures the e2e flows from a Data Center orchestrator perspective.
OCP GPU & Accelerator Management Interfaces: This document describes industry standard formats and protocols that make it easier for CSPs (Cloud Service Providers, hyperscalers) to onboard new GPU and accelerator designs with less toil and faster time to market while reducing manageability permutations for suppliers.
OCP GPU & Accelerator RAS Requirements: This document standardizes certain RAS (Reliability, Availability, and Serviceability) capabilities for GPU and accelerator devices. The primary goals are to (a) ensure that GPU/accelerator devices are sufficiently reliable, available, and serviceable to meet hyperscale data center requirements; (b) increase interoperability by standardizing interfaces and mechanisms for certain features; and (c) allow data center operators to validate firmware and software stacks and verify end-to-end error handling and resiliency workflows.
Requirements
| Type | Title | Revision | Date |
|---|---|---|---|
| Specification | OCP GPU FW Update Specification v1.1 | 1.1 | November 7, 2025 |
| Specification | OCP GPU FW Update Specification v1.0 | 1.0 | October 1, 2024 |
| Specification | OCP GPU FW Update Specification v0.7 | 0.7 | June 1, 2023 |
| Type | Title | Revision | Date |
|---|---|---|---|
| Specification | OCP GPU & Accelerator Management Interfaces v1.1 | 1.1 | November 7, 2025 |
| Specification | OCP GPU & Accelerator Management Interfaces v1.0 | 1.0 | October 1, 2024 |
| Specification | OCP GPU & Accelerator Management Interfaces v0.5 | 0.5 | June 1, 2023 |
| Type | Title | Revision | Date |
|---|---|---|---|
| Specification | OCP GPU & Accelerator RAS Requirements v1.7 | 1.7 | November 7, 2025 |
| Specification | OCP GPU & Accelerator RAS Requirements v1.0 | 1.0 | October 1, 2024 |
| Specification | OCP GPU & Accelerator RAS Requirements v0.9 | 0.9 | August 1, 2024 |
| Specification | OCP GPU & Accelerator RAS Requirements v0.5 | 0.5 | June 1, 2024 |
The compliance tool (CTAM) is available here https://github.com/opencomputeproject/ocp-diag-ctam