From Project to Production: HAMi and Viettel Cloud at KCD Vietnam

Kubernetes treats GPUs as atomic resources, forcing over-provisioning and low utilization in multi-tenant AI Notebooks. HAMi's vGPU virtualization and DRA solve this, but only if implemented correctly. This talk at KCD & OpenInfra Days Vietnam covers the mechanics of GPU sharing (DRA resource requests, HAMi fractional GPU allocation, memory isolation, compute slicing) and the production deployment at Viettel Cloud: architecture, bottlenecks, and operational realities of fractional GPUs for data science workloads at telco scale.