Where to Process Camera Video: On-Premise or in the Cloud?
Camera video can be processed on a company server, at a remote compute site or through a hybrid arrangement. The choice depends on data rules, connectivity and load. At Busted, we select architecture after defining tasks and video sources.
On-site processing
In an on-premise setup, cameras send video to a computer or server inside the business. Analysis and event generation run there. If managers need remote report access, define separately what data leaves the site and how.
An on-site server needs a suitable GPU, CPU, RAM, storage, power, cooling and networking. Size it for the actual task mix. Establish who maintains it and how the system recovers after a stoppage.
Remote processing
In a remote setup, streams or selected data go to a compute server. An intermediary on-site computer may be needed to receive camera video and transmit it over an agreed connection. Test bandwidth, connection stability and outage behavior.
Decide whether to retain a full archive or only events, frames and short clips. This affects transfer volume, storage and later review.
Selecting a GPU
Camera count explains only part of the load. Resolution, analysis frequency, selected models, scenario count and video operations matter. NVIDIA DeepStream benchmarks assess the whole pipeline, including decoding, preprocessing, inference and output generation.
Validate GPU sizing with selected streams. Check resource utilization, latency and headroom for normal fluctuations. Assess storage and video display separately.
Choosing a setup
Define processing-location requirements, report availability and acceptable latency. Then compare options by total cost, maintenance and resilience.
Busted helps select equipment and organize processing within the agreed environment. Start with a camera description and one task that needs a practical architecture.