Surveillance & Security
Recording, analytics and face recognition hardware built for the camera estate you actually run
- VMS ingest per recording server: 64+ cameras at 8MP
- Face recognition channels: Up to 200 simultaneous
- Face template database: Up to 1 million templates
- Storage density for retention: 60-bay 4U JBOD
The challenge
A camera estate is judged on the days you need footage that turns out not to be there, or the frame that a face-recognition system was never sized to match. Recording servers, GPU analytics nodes and long-term storage each have different failure modes under 24x7 load, and a general-purpose server sized for office workloads will find all three eventually. Government, railway and large-campus deployments add assessment regimes and procurement processes on top of that, which most hardware catalogues are not built around.
Our approach
Spectra builds three purpose-designed NetBytes platforms rather than one generic box wearing a surveillance label: 2U recording servers for VMS ingest, GPU-dense analytics servers, and a face-recognition appliance sized to 200 channels and a million-template database. Each is validated against the customer's own VMS choice — Milestone, Genetec, DSSL TRASSIR and others — and OEM relationships with Videonetics and i2V mean an analytics stack can ship configured with the hardware instead of being integrated on site afterwards. Storage is sized against actual retention math using surveillance-rated drives, not desktop drives repurposed for 24x7 write loads. Everything is integrated and burn-in tested in India, and new firmware and GPU driver combinations are proven in the company's own data centre before they go into a customer build.
What makes this workload hard
- Retention math gets guessed, not calculated Camera count, resolution, codec and retention days interact in ways that are easy to under-provision for, and the shortfall only shows up when footage is needed.
- Analytics bolted on after the fact Adding number-plate recognition or crowd analysis to a recording server that was never sized for GPU load usually means a second box, not an upgrade.
- Assessment regimes that generic hardware ignores Government and railway procurement runs on STQC and RDSO assessment, and a server line built for a global catalogue rarely accounts for that from the start.
- Vendor lock-in through the back door Hardware tuned for one VMS vendor's reference design makes switching VMS later an expensive re-platforming exercise, not a software change.
- Face recognition at scale changes the sizing problem Matching against a database of hundreds of thousands of templates in real time needs GPU and memory sizing that camera-counting alone does not capture.
- 24x7 write loads on drives never meant for it Desktop and nearline drives degrade under continuous surveillance write cycles in a way that only becomes visible as failures, months into a deployment.
What we deliver
- VMS Recording Servers 2U chassis rated for 64+ cameras at 8MP with GPU-accelerated analytics headroom built in, tuned for sustained write throughput rather than burst performance.
- Face Recognition Appliance 2U appliance supporting up to 200 simultaneous channels against a database of up to one million templates, with GPU scheduling tuned for low-latency matching.
- Retention Storage 12-bay 2U and 60-bay 4U JBOD storage for long-term retention, built around surveillance-rated drives designed for continuous write cycles.
- Vendor-Agnostic Builds Platforms are validated against the requirements of the VMS you already run — Milestone, Genetec, DSSL TRASSIR and others — rather than tying you to one ecosystem.
- Analytics OEM'd In OEM relationships with Videonetics and i2V mean video-analytics software can be supplied and supported alongside the hardware as one line, not integrated after delivery.
- GPU-Accelerated Analytics NVIDIA GPUs sized to your stream count and model complexity, from a couple of GPUs for a single site to dense multi-GPU nodes for city-scale analysis.
The hardware: Surveillance server hardware