HAMi vs Kueue
HAMi
project-hami.ioKueue
kueue.sigs.k8s.ioSide by side
| Figure | HAMi | Kueue |
|---|---|---|
| Price from | Free | Free |
| Free plan | Yes | Yes |
| Free trial | — | — |
| Open source | — | — |
| Linux | Yes | — |
| API | — | Yes |
| Self-hosted | Yes | Yes |
| Plans published | 1 | 1 |
| Spec score · GPU Cluster Management Software | 8.3 (#5) | 8.0 (#7) |
| Community support | The project links to community support through Discord and Slack (#hami-dev). | The project directs users to Kubernetes Slack, GitHub for bugs and contributions, and a mailing list for announcements. |
| Monitoring | HAMi provides allocation counts and spread plus real-time GPU memory and core utilization visibility. | Kueue exposes Prometheus metrics for system health and ClusterQueue and LocalQueue status, with Grafana query guidance in its docs. |
| Purpose | HAMi is open-source, cloud-native GPU virtualization middleware for sharing, isolating, and scheduling heterogeneous accelerators on Kubernetes. | Kueue manages quotas and job resource consumption in Kubernetes, deciding when jobs wait, start, or are preempted. |
| Scheduling | HAMi offers binpack, spread, and topology-aware scheduling policies. | Kueue supports priority-based queueing, resource flavor fungibility, preemption, and all-or-nothing scheduling. |
Plans
In their own words
HAMi
HAMi is open-source middleware for sharing and scheduling GPUs and other accelerators among AI workloads on Kubernetes. Workloads can request GPU memory and core limits through Kubernetes resource controls. Scheduling options include binpack, spread and numa-first, with binpack and spread available at node or GPU scope. The supported-device list spans NVIDIA, AMD, AWS Neuron, Huawei Ascend, Cambricon, Enflame, Hygon, Iluvatar, Kunlunxin, MetaX, Moore Threads, Vastai and Biren; isolation and partitioning capabilities vary by device. HAMi works with Kubernetes APIs, DRA and CDI, and documents scheduling integrations including Volcano, Kueue, Koordinator and NVIDIA KAI Scheduler. Its node metrics endpoint exposes device memory and core utilization for Prometheus. Classic HAMi requires Kubernetes v1.23 or later. HAMi-DRA requires Kubernetes v1.34 or later, the DRA Consumable Capacity feature gate, CDI and NVIDIA driver 440 or later. NVIDIA limits use user-space library interception, which does not cover applications that bypass CUDA. Enforcement is described as soft and best-effort; the FAQ recommends MIG when hardware-enforced isolation is required.
Kueue
Kueue is a free Kubernetes project that manages quotas and job resource consumption, deciding when workloads wait, start, or are preempted. It installs on a standard Kubernetes cluster without replacing existing components. Its scheduling tools include priority-based queues, fair sharing, quota borrowing between teams, resource flavor fungibility, preemption policies, and all-or-nothing scheduling. Built-in integrations cover Kubernetes Jobs and Pods, Kubeflow training jobs, Ray jobs and clusters, JobSet, AppWrapper, Deployments, StatefulSets, and LeaderWorkerSet. MultiKueue can find capacity and dispatch workloads across clusters, while topology-aware scheduling accounts for data center layout to improve communication throughput between pods. Kueue provides Prometheus metrics and an endpoint for viewing pending workloads. Installation instructions cover kubectl, Helm, and source builds. The guide recommends Kubernetes 1.34 or newer and officially supports and tests only those versions. Kueue also provides administrator and user ClusterRoles, with resource-specific roles that can be composed into custom roles. Its webhook server includes internal certificate management, and the installation guide links to cert-manager instructions.