How Intsig Serves Billions of Document Scans: GPU Virtualization at...
Talkat KubeCon + CloudNativeCon + OpenInfra Summit + PyTorch Conference China 2026 · 2026-09-07..09-09·
summary
Explains how Intsig reduced queue times and raised GPU utilization for massive-concurrency OCR workloads by switching from Tencent QGPU to HAMi (a CNCF Sandbox project) for GPU virtualization, scheduling, and monitoring across ~1,000 GPU cards. Describes production results and patterns from other large deployments (SF Express, China Merchants Bank, NIO, ICBC), including cost/efficiency lessons such as why slicing GPUs below 1/6 is harmful, and ends with a demo of Chaterm, an open-source AI terminal for GPU cluster operations.
AI-written summary · about
related talks
28mShared Yet Isolated at Scale: Building Multi-Tenant Inference Platform on…
38mStop Allocating GPUs, Start Delivering Intelligence: An Enterprise...
07mYour GPU delivery is here! Now what? A 5 minute guide to make your CTO happy
28mGPU sharing done right: Secrets, security, and scaling with Vault and vCluster
27mUnlocking Heterogeneous AI Infrastructure K8s Cluster: Leveraging the Power of HAMi
09mLightning Talk: Bringing Multi-Tenancy to GPU-Based Workloads
17mKeynote
32mSmart GPU Management: Dynamic Pooling, Sharing, and Scheduling for AI Workloads in Kubernetes