Einblicke, Tutorials und Neuigkeiten von unserem Team
At KCD Beijing 2026, HAMi core contributors presented the GPU scheduling paradigm shift from Device Plugin to DRA. This post reviews the key insights — DRA's capabilities and challenges, HAMi-DRA's automated Webhook approach to lowering migration barriers, and real-world performance and observability results.
This weekend, the Dynamia team heads to Amsterdam for KubeCon + CloudNativeCon Europe 2026. As the initiator and leading contributor of the CNCF Sandbox project HAMi, Dynamia will participate in multiple core activities representing the HAMi community.
A systematic upgrade to HAMi's website and docs, improving community visibility, content structure, search, and usability.
Ke Holdings Inc. achieves nearly 3x improvement in GPU utilization (13% → 37%) by implementing AIStudio, a smart computing platform built on Kubernetes and HAMi for GPU virtualization across multi-cloud environments.
Dynamia and OpenCSG announce a strategic partnership to build a compute-native foundation for enterprise AI workloads. By natively integrating HAMi's heterogeneous GPU virtualization and scheduling capabilities into CSGHub, both parties will advance model platforms from "model hosting" to "compute-native."
A look back at how HAMi evolved from a single GPU virtualization solution into a unified platform supporting 11+ AI accelerator vendors.
HAMi v2.8 deep dive: Kubernetes DRA support, Leader Election mechanism, CDI mode support, upstream ecosystem integration (Kueue, vLLM), critical bug fixes, and stability improvements.
As AI inference and training move toward multi-model co-location and fragmented concurrency, compute virtualization is evolving from local optimization to a critical infrastructure capability. Based on public repositories and community-visible information, this article analyzes Flex:ai's current verifiable delivery boundaries from an engineering perspective, and discusses the common engineering gaps as compute virtualization projects move from 'working' to 'dependable infrastructure'.
Against the backdrop of sustained growth in AI training and inference demand, DaoCloud uses CNCF Sandbox open-source project HAMi in its public and private GPU cloud platforms to build a more flexible, cloud-native GPU resource management approach.
Get the latest technical articles, tutorials, and HAMi community updates.