Skip to navigation
Appendix

Performance

View as Markdown

This section documents benchmark results for VSS agent workflows and Real-Time Video Intelligence microservices. The results include measurements from VSS 3.2 and 3.3; each linked page identifies its own tested release, model, precision, and hardware. Use those page-specific test configurations when sizing GPU infrastructure or comparing use cases.

Agent Workflows

Agent WorkflowPageWhat Is Measured
Alert Verificationperformance-alert-verification- End-to-end latency and verified alert throughput for live RTSP alert verificationRT-DETR and GDINO with CR3 across RTX PRO 4500, RTX PRO 6000, DGX H100 SXM, DGX Spark, and AGX ThorVLM precision (BF16, FP8, NVFP4) and stream-concurrency scaling characteristics
Video Summarizationperformance-lvs- E2E latency vs. video length (1 min–12 hr)Burst max concurrency at target avg latency (1, 5, 10 min videos)Cosmos3 Nano Reasoner FP8 across H100, RTX Pro 6000 SE, and L40S
Searchperformance-search- E2E mean and P90 latency vs. concurrent query load (1–50)Per-stage latency breakdown on H100 and RTX Pro 6000 SEVideo ingestion latency

Microservices

MicroservicePageWhat Is Measured
Real-Time CV (RT-CV)performance-rt-cv- Maximum concurrent 1080p streams per GPURT-DETR (Resnet50 and EfficientViT/L2) and Grounding DINO modelsLatency-sensitive and throughput-maximizing operating modes
Real-Time VLM (RT-VLM)performance-rt-vlm- Maximum concurrent streams and chunk latency for live RTSP streamsTotal latency and throughput for pre-recorded video file processingAlerting (OSL=1) and captioning (OSL=100) use casesCosmos3 Nano Reasoner via vLLM
Real-Time Embedding (RT-Embedding)performance-rt-embed- Maximum concurrent streams for live RTSP video embeddingVideo and text embedding throughput and latency in request modeFile embedding latency vs. video lengthCosmos-Embed1, FP16, TensorRT
VIOSperformance-vios- RTSP and WebRTC stream profiling (GPU/CPU utilization)Max concurrent streams per replica on L40 and H100Single-stream and concurrent video-download latency (H.264 and H.265)Single-stream and concurrent picture API latencyRTX PRO 6000 Blackwell benchmarks with and without transcoding