Insights, tutorials, and news from our team
On July 25, the CMB Digital Intelligence Innovation Meetup landed in Hangzhou. HAMi community Approver and Dynamia engineer Wang Jifei delivered a talk on "Making NPU Easier to Share." This article reviews the full evolution of HAMi vNPU from hard to soft partitioning, and its joint validation in production with China Merchants Bank.
On July 2, 2026, HAMi, the open-source project founded and long led by Dynamia, officially became a CNCF Incubating project. As the only open-source project today focused on heterogeneous GPU virtualization and efficient scheduling that has reached the CNCF Incubating stage, HAMi's milestone sends a clear signal: the AI infrastructure race is shifting from "who has more compute" to "who can manage compute more efficiently."
In the industry window opened by HAMi's graduation to CNCF Incubating, Dynamia organized and accompanied a CNCF and PyTorch Foundation delegation on visits to leading Chinese AI chip and compute infrastructure companies—including Biren, Iluvatar CoreX, MetaX, Cambricon, and Super Cloud—to discuss adaptation, ecosystem co-building, and production deployment between domestic AI accelerators and the global cloud native AI Infra open-source ecosystem.
A CNCF case study on how China Merchants Bank built a unified AI compute scheduling platform on Kubernetes and HAMi, enabling heterogeneous resource pooling, topology-aware scheduling, and fine-grained accelerator sharing. Hardware pool utilization reached 100%, cross-machine scheduling dropped 30%, and partitioning granularity went down to 1 GB memory / 1% compute.
On July 2, 2026, HAMi officially graduated to a CNCF Incubating project. Following its acceptance as a Sandbox project in August 2024, this milestone marks HAMi's growth from a promising newcomer into trusted infrastructure widely adopted in real production environments.
KAI Scheduler originates from the core scheduling engine of Run:ai, was open-sourced by NVIDIA, and has entered the CNCF Sandbox. Led by Dynamia, the HAMi-core integration has now landed in the KAI Scheduler mainline, giving its GPU sharing capability hard runtime memory isolation.
After the landmark Keynote Demo at KubeCon EU 2026, the HAMi community is heading to KubeCon India 2026. CNCF Ambassador Saiyam Pathak will deliver a keynote featuring a live GPU sharing demo on NVIDIA DGX Spark.
A community contributor's deep dive into diagnosing and fixing vGPU scheduling delays (from seconds to 10+ minutes) after upgrading a 200-node GPU cluster from Volcano 1.7 to Volcano 1.12 + hami-dp. Root cause analysis of API Server throttling, patch delays, and resource view timeouts.
A hands-on guide covering HAMi, DRA-mode deployment, GPU memory and compute partitioning, native DRA usage, and DevicePlugin-compatible migration.
Get the latest technical articles, tutorials, and HAMi community updates.