Ecosystem Resources
Many cloud providers, hardware vendors, and platform builders maintain their own Kueue documentation, tutorials, and reference implementations, and some ship Kueue as a built-in component of their AI/HPC stacks.
This page collects those materials and third-party ecosystem integrations in one place, so that you can find the guidance relevant to your environment. For community tutorials, conference presentations, and adopter stories, see Talks, Case Studies & Tutorials.
Note
The resources below are owned and maintained by the respective vendors, not by the Kueue project. They may target specific Kueue versions or vendor-customized builds, and can drift from the upstream documentation. Always cross-check version compatibility with the official docs and the vendor’s release notes. Listing here is not an endorsement; the order is alphabetical and the list is not exhaustive — contributions adding further materials are welcome.Amazon Web Services (EKS, SageMaker HyperPod)
Amazon SageMaker HyperPod task governance is built on top of Kueue: it manages quotas, priorities, preemption, and gang scheduling for EKS-based HyperPod GPU clusters through Kueue APIs.
Official documentation:
- SageMaker HyperPod task governance — overview and setup of the Kueue-based governance layer.
- Using gang scheduling in Amazon SageMaker HyperPod task governance — all-or-nothing admission for distributed training.
- Using topology-aware scheduling in Amazon SageMaker HyperPod task governance — network-topology-based placement.
Best practices and reference architectures:
- Best practices for Amazon SageMaker HyperPod task governance — quota design and team sharing patterns.
- Maximize HyperPod Cluster utilization with HyperPod task governance fine-grained quota allocation — borrowing and lending compute across teams.
- Schedule topology-aware workloads using Amazon SageMaker HyperPod task governance — walkthrough of topology-aware placement on HyperPod.
- Introducing AI on EKS: powering scalable AI workloads with Amazon EKS — AWS open-source initiative for AI on EKS, including Kueue-based scheduling blueprints.
- MLOps on Amazon EKS - Comprehensive Guide — MLOps reference stack on EKS with Kueue as a modular component.
CoreWeave (CKS)
- Kueue on CoreWeave Kubernetes Service — CoreWeave-maintained Helm chart, with metrics automatically ingested into a Kueue Grafana dashboard.
- Kueue: A Kubernetes-native System for AI Training Workloads — how AI labs use Kueue on CKS for training and batch inference.
Google Cloud (GKE, AI Hypercomputer)
Getting started, quotas, and prioritization:
- Create an AI-optimized GKE cluster — AI Hypercomputer cluster creation, with Kueue in the workload-scheduling toolchain.
- Deploy a batch system using Kueue — the canonical GKE Kueue starter tutorial (two tenant teams sharing a cluster).
- Implement a Job queuing system with quota sharing between namespaces on GKE — cohorts and quota borrowing on GKE.
- Best practices for running batch workloads on GKE — platform-level guidance with Kueue at the center of the queueing story.
- Best practices for GKE AI/ML workload prioritization — priorities and preemption for AI workloads.
- Optimize GKE resource utilization for mixed AI/ML training and inference workloads — sharing cluster capacity between batch training and inference workloads.
Provisioning, topology, and accelerators (GPUs & TPUs):
- Run a large-scale workload with flex-start with queued provisioning — Kueue-managed
ProvisioningRequestadmission on GKE. - Optimize AI training on TPUs with DWS, Ray and Kueue — Dynamic Workload Scheduler with KubeRay and Kueue.
- Schedule GKE workloads with Topology Aware Scheduling — Kueue TAS on AI-optimized GKE clusters.
- Orchestrate Multislice workloads using JobSet and Kueue — TPU Multislice training with JobSet and Kueue.
- Schedule dynamic slices with Kueue and TAS — Ironwood (TPU7x) dynamic slicing on GKE, using Kueue TAS together with a dedicated slice
AdmissionCheckcontroller that creates the Slice custom resources and holds admission until GKE activates the Slice. - Enable node health prediction — configuring GKE AI Hypercomputer node health prediction with Kueue.
Multi-cluster, observability, and reference blueprints:
- Kueue Arrives in the Google Cloud Console — native GKE Console UI for monitoring Kueue queues, quotas, and pending workloads.
- With MultiKueue, grab GPUs for your GKE cluster, wherever they may be — using MultiKueue and DWS across many Google Cloud regions.
- MultiKueue with ClusterProfile API and GKE Fleet — end-to-end walkthrough of using MultiKueue with GKE Fleet.
- Google Cluster Toolkit (formerly HPC Toolkit) — blueprints for AI/HPC clusters on Google Cloud, including Kueue with TAS.
- GKE AI Labs — reference architectures, tutorials, and blueprints for AI/ML on GKE.
- NeMo Framework on Google Kubernetes Engine (GKE) to train Megatron LM — end-to-end Megatron-LM training recipe using Kueue on GKE.
IBM
- MLBatch — queueing and quota management setup for AI/ML batch jobs, combining Kueue, AppWrapper, Kubeflow Training Operator and KubeRay, with detailed operational best practices.
- AppWrapper — the AppWrapper controller for Kueue.
- Improve GPU utilization with Kueue in OpenShift AI. How IBM achieved 90% GPU allocation in Vela — production deployment of Kueue on IBM’s Vela AI supercomputer.
Microsoft Azure (AKS)
- Install and Configure Kueue on Azure Kubernetes Service (AKS) — AKS-maintained overview and installation guide.
- Schedule and deploy batch jobs with Kueue on Azure Kubernetes Service (AKS) — ResourceFlavors, ClusterQueues and sample batch jobs on AKS.
- AI Conformant Azure Kubernetes Service (AKS) clusters — AKS engineering blog positioning Kueue within the Kubernetes AI conformance stack.
Oracle Cloud Infrastructure (OKE)
- Running RDMA (Remote Direct Memory Access) GPU Workloads on OKE — Oracle’s reference stack for GPU clusters on OKE; recent versions deploy Kueue (with Topology, ResourceFlavor, ClusterQueue, and LocalQueue objects) by default.
- Using RDMA Network Locality When Running Workloads on OKE — configuring topology-aware RDMA placement with Kueue on OKE.
Red Hat
- Red Hat build of Kueue documentation — official documentation for installing, configuring, and using the Red Hat build of Kueue on OpenShift.
- Red Hat OpenShift is joining the Kueue — overview of Red Hat build of Kueue on OpenShift.
- Improve GPU utilization with Kueue in OpenShift AI. How IBM achieved 90% GPU allocation in Vela — using Kueue in Red Hat OpenShift AI for multi-tenant GPU clusters.
Other external docs
Workload frameworks:
- Ray: Gang scheduling, Priority scheduling, and Autoscaling for KubeRay CRDs with Kueue — official KubeRay guide for orchestrating
RayJob,RayCluster, andRayServiceworkloads with Kueue. - LeaderWorkerSet: Topology Aware Scheduling with Kueue — configuring Kueue Topology Aware Scheduling for multi-host
LeaderWorkerSetinference workloads.
UI and observability:
- Headlamp: Kueue plugin — web UI plugin for inspecting Kueue Workloads, LocalQueues, ClusterQueues, ResourceFlavors, and Cohorts (see the announcement and troubleshooting walkthrough).
Feedback
Was this page helpful?
Glad to hear it! Please tell us how we can improve.
Sorry to hear that. Please tell us how we can improve.