81 HD streams, 30 fps of inference on each one, in a single on-premise server — under 40 W of accelerator power.
White paper
Advanced Analytics and AI Accelerators in VMS
VMS is no longer about displaying and storing video. It is becoming an intelligent platform — actionable insight, real-time alerts, automated response.
As video volume grows, traditional VMS struggles to keep up with the demand for real-time processing. AI analytics turns raw footage into information: anomalies detected, patterns recognised, alerts generated, while the operator is still watching wall one of twelve.
Rising demand for security and surveillance, smart city programmes, and commercial and industrial expansion — all of it landing on the VMS layer.
Machine learning applied to live video automates work that was previously impossible, or needed a human watching every frame.
Faces, gait, clothing. Match against a database of known individuals to prevent unauthorised access and flag persons of interest in airports, government buildings, offices and warehouses.
Vehicles, bags, packages — identified and followed automatically. Central to perimeter security, where unauthorised intrusion has to be caught as it happens, not found in playback.
Loitering, crowd formation, sudden movement. Interpreting what people and groups are doing lets the system raise an alert and enable pre-emptive action before an incident.
LPR / ANPR for traffic management, tolling, parking enforcement and access control — plates read and recorded automatically, stolen vehicles identified as they pass.
Movement and density of people or vehicles, measured in real time — large events, traffic flow optimisation, public safety decisions made while there is still time to make them.
AI processing needs specialised hardware, and it can sit at three points in the video pipeline. Each choice buys something and costs something.
A balanced VMS should put AI processing in the on-premise edge server — lower operational cost, easier AI updates and maintenance, more functionality, and compliance with data privacy regulation.
MemryX, Advanced Analytics and AI Accelerators in VMSThe same three options every VMS integrator weighs — and what each one actually costs to run once the cameras are live.
Traditionally manages the flow of video through the VMS. Cost-effective for moving data — not cost-effective for running the analytics on it.
Parallel processing gives a real jump over CPUs for inference. It also brings high energy consumption, higher operating cost, and heat that usually needs dedicated cooling.
Silicon built only for inference. Balances performance and efficiency, drawing significantly less power than a GPU at comparable or better throughput — which is what makes edge deployment practical.
Four MemryX M.2 modules on a PCIe carrier card, decoding and inferencing HD video in one on-premise server.
YOLOv5s at 320×320, 30 fps inference per video stream. Four MemryX M.2 modules on a PCIe carrier card, under 40 W total AI accelerator power. AMD x86 Ryzen host used for the demonstration.
Derived from over 100 different versions of the YOLOv5 and YOLOv8 object detection models, varying in layer depth, input resolution and complexity. Batch=1, Intel Core i7-13700, 64 GB RAM.
Across a set of 50 different AI vision model benchmarks published on the competitor’s own website. Models compiled with each company’s SDK and measured on similar x86 host platforms.
An accelerator designed for exactly this workload: high performance, low power, and integration that does not mean rebuilding the system around it.
Processing power to handle real-time analytics at the edge, so the VMS responds to threats and anomalies as they happen.
Less power means lower operating cost and less heat — deployable in environments with limited cooling infrastructure.
Supports a wide range of models and frameworks, and drops into an existing VMS without overhauling the infrastructure around it.
Deploy in the configuration the site needs today and add modules as the camera count and the model set grow.
A smart city project reads thousands of camera feeds in real time — anomalies detected, objects tracked, insight handed to city officials — and stays cost-effective and scalable as the surveillance estate grows.
That is the shape of the MemryX advantage in VMS: high performance, low power draw and easy integration together, so advanced analytics arrive without a trade against cost or complexity.
Talk to our team about MX3 in your VMS deployment, or start today in the Developer Hub — the full toolchain is a free, ungated download.
Cascade modules ship through our distribution partners. Pick the channel that suits your region and order volume.
