Container × GPU × AI Pod Platform

AI Computing Platform – VIDAware Containerization

A flexible runtime and real-time scheduling hub centered on AI service units (Pods)

Centered on an AI containerized computing platform, each AI model, video analytics, and event inference service is packaged as an independent AI service unit (Pod). According to real-time needs across city sites, services are automatically assigned to different GPU computing nodes, enabling real-time scaling, uninterrupted operation, and cross-node fault tolerance for city-scale AI computing.

Fragmented Compute

Idle resources / Unmanaged

VIDAware Compute Containerization System Architecture

Elastic Container × GPU × AI Pod Architecture

VIDAWARE × AI Pod Orchestration

Multi-Source Streams → AI Pod → Dynamic GPU Resource Scheduling

Input Streams

Primary Pod

Standby Pod

Sources are automatically dispatched → Pod scheduling → lowest-GPU-load priority

GPU Server1

CUDA

DEEP_LEARNING(Deep Learning)

18%

GPU Server2

CUDA

GENERAL_COMPUTE(General Compute)

29%

GPU Server3

CUDA

EDGE_INFERENCE(Edge Inference)

44%

GPU Server4

CUDA

VIDEO_TRANSCODE(Video Transcoding)

57%

GPU Server5

CUDA

MODEL_SERVING(Model Serving)

68%

GPU Server6

CUDA

BATCH_INFERENCE(Batch Inference)

83%

Management & Operations Layer: Commercialization and Security

Ensures data security among tenants and supports a computing-token-based business model.

Multi-Tenant Compute Token Billing

Tracks resource usage for each project and converts it into token deductions. The chart shows token consumption by major tenants across different services.

Face Recog License Plate Event Detect

Authentication and Security

SSO / OIDC, API key management, and end-to-end TLS encryption.

Multi-Tenant Isolation

Namespace + RBAC ensures complete resource isolation between tenants.

Real-Time Alert System

Automatic alerts via email / LINE / SMS.