ClearML vs HAMi

ClearML is an AI infrastructure platform organized into an Infrastructure Control Plane, AI Development Center, and GenAI App Engine. The Control Plane manages GPU resources across on-premises, cloud, and hybrid environments, including multi-tenant GPU-as-a-Service. The Development Center supports developing, training, testing, and deploying AI/ML models, with monitoring, pipelines, a model repository, and CI/CD integration. The GenAI App Engine deploys LLMs on compute clusters and provides networking, authentication, access control, and monitoring. Experiment tracking, comparisons, artifacts, metrics and plots, and Git version control integration appear across the listed tiers. Integrations include PyTorch, TensorFlow, Keras, scikit-learn, XGBoost, LightGBM, Optuna, Hydra, TensorBoard, Matplotlib, and LangChain; storage integrations include AWS S3, Azure, Google Storage, MinIO, and Backblaze B2. ClearML can be hosted on its servers, self-hosted, or provided as a managed service, with VPC, on-premises including air-gapped, and hybrid options. Community is free; Pro costs $15 a month per user plus usage. Scale and Enterprise pricing is by quote.

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.