Einblicke, Tutorials und Neuigkeiten von unserem Team
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.
GPU management in research labs has always been a headache. Boardware spent three years evolving from VM-exclusive access to a Kubernetes + HAMi cloud-native solution, ultimately achieving physical isolation of Agent RL training and inference on a 12-GPU cluster. This is the seventh article in the HAMi Community Meetup Shenzhen recap series.
Sangfor Cloud AI Chief Architect Jia Haojie shares a complete enterprise AI computing governance practice — from the computing challenges of the Agent era to intelligent routing and safety guardrails of the AI Computing Gateway, to Volcano + HAMi production-grade vGPU scheduling.
HAMi v2.9.0 is officially released! Ascend 910C user-space virtualization (HAMi-core mode) delivers fine-grained memory and compute sharing, HAMi-DRA reaches production readiness based on Kubernetes DRA standards, Vastai device support is added, and security and observability are comprehensively enhanced.
A CNCF case study on how SNOW Corp., a NAVER subsidiary, uses HAMi for GPU sharing and KEDA for proactive autoscaling to handle 700% viral traffic spikes — achieving 91% MTTR reduction, 85% fewer surge errors, and USD 17.4M in estimated cost savings.
China Merchants Bank built a unified heterogeneous compute infrastructure on HAMi, achieving super-node dual-chip module adaptation (100% GPU pooling rate), vNPU-Core software partitioning (1 GB VRAM / 1% compute granularity), and three-tier topology-aware scheduling (30% reduction in cross-node scheduling). The solutions have been contributed back to the HAMi open-source community.
Chen Junchao from SF Technology shares the complete evolution from bare-metal to a multi-cloud hybrid architecture, with an in-depth look at HAMi-based GPU pooling, memory oversubscription, and mixed deployment in production.
HAMi Maintainer Li Mengxuan reveals v2.9 core features for the first time at the Shenzhen Meetup, deep-diving into the DRA technology roadmap and the community vision for unified heterogeneous compute management.
Get the latest technical articles, tutorials, and HAMi community updates.