About Codex GPU Queue
Codex GPU Queue is a local Windows tool that lets multiple Codex sessions share a single NVIDIA GPU. Jobs submitted from different sessions queue in the background, start when resources become available according to scheduling rules, and return results to their originating sessions. The tool is open-source and was released as a documented version of the maker's daily-use scheduler.
Review
Codex GPU Queue addresses a specific friction point for developers running several concurrent Codex sessions on one Windows machine. Instead of manually deciding which session gets the GPU next, you submit jobs and the broker handles the rest. The tool doesn't aim to be a full cluster scheduler-it's a local queue for a single GPU, built from a real workflow and released for others to try.
Key Features
- Background job queue that starts submitted GPU work when resources and scheduling rules allow
- Results automatically return to the originating Codex session, removing the need for manual coordination
- Read-only waiting-reason observer that reports confirmed blockers from a snapshot, keeps uncertain reasons undetermined, and omits commands, private paths, and raw job identities
- Tested with 51 CPU/synthetic tests passing in a clean dependency environment on the same Windows machine
- Demonstrated with two independent Codex sessions submitting synthetic CUDA jobs-B queued while A ran, then started automatically after A completed
Pricing and Value
Codex GPU Queue is free and open-source. The pricing page doesn't describe any paid tiers, subscriptions, or licensing fees. The release is a source candidate for trusted local workflows. An optional Node/SDK continuation controller is mentioned as not included in this release, but no pricing information is attached to it.
Pros
- Eliminates the manual task of deciding which Codex session gets the GPU next
- Open-source release lets Windows AI builders inspect, modify, and run the tool locally
- Waiting-reason observer gives a snapshot view of blockers without exposing sensitive paths or job details
- Verified with synthetic tests and a real two-session demo, not just described in documentation
- Lightweight-runs in the background on a single machine without external dependencies
Cons
- Limited to one Windows machine and one NVIDIA GPU; multi-GPU or distributed setups aren't supported
- No runtime model API-the demo uses synthetic CUDA computation, not actual model inference
- Not well suited for users who need a managed cloud queue, cross-platform support, or built-wheel installation paths
Codex GPU Queue fits developers who already run multiple Codex sessions on a single Windows box with one NVIDIA GPU and want to stop manually juggling GPU access. The tool's scope is deliberately narrow-it doesn't handle model inference directly or span multiple machines. For that specific local workflow, it replaces a repetitive coordination task with a background process that runs jobs when they're eligible.
Open 'Codex GPU Queue' Website
Your membership also unlocks:
