Open Compute Project
Work I've contributed to through OCP, mostly around industry standards for AI infra debug, RAS, and firmware updates, keeping them consistent across silicon vendors instead of bespoke per supplier.
-
Hyperscale CPU Debug and RAS Requirements Specification
Standardizes how CPUs across vendors expose debug data (crashdumps, extended register dumps, trace logs, and state dumps), plus a common workflow for collecting them in production. Also lays groundwork for standardized diagnostics packaging so hyperscalers aren't building one-off tooling per silicon vendor. Co-authored with engineers from Google, AMD, Arm, Meta, and Intel under OCP.
-
Hyperscale CPU Impactless Firmware Updates Requirements Specification
Defines requirements for updating CPU/SoC firmware (microcode, BIOS, management-mode firmware) in-band or out-of-band, without a reboot or service disruption. Aimed at closing the gap between "we found a critical security fix" and "we can ship it fleet-wide without downtime." Co-authored with engineers from Google, AMD, Arm, Meta, and Intel under OCP.
-
CPACT (OCP Diagnostics Compliance Tool)
An open-source compliance framework that lets CPU suppliers self-verify their implementation against the debug, telemetry, and impactless-update specs above, before a hyperscaler ever tests it. Supports multiple access paths (in-band Linux, Redfish, BMC) and defines compliance scenarios in YAML so testing scales across vendors instead of being bespoke per supplier.