System administrator, integrator, purchasing
Server requirements and hardware calculations
The minimum for a pilot and recommended resources for 100 cameras, 100 sensors and 10 access control controllers.
- 1
Not just the number of cores - you need core power
16 cores at 1 GHz will not replace 8 cores at 3 GHz. For EAS, frequency (base ≥2.0–2.4 GHz) and CPU generation (Core i5/i7 8+, Xeon Silver/Gold, Ryzen/EPYC) are important. Atom, Celeron ≤1.5 GHz and cloud vCPUs with burst without a guaranteed share are not suitable. Benchmark: CPU Mark (multi) from ~6,000 (pilot) to ~25,000+ (100 cameras) - cpubenchmark.net.
- 2
Minimum - just for starting
Pilot: 2 cores **≥2.0 GHz**, 4 GB RAM, 80 GB SSD - up to 8 cameras. Not Atom or “2 vCPU 1 GHz”.
- 3
Profiles by scale
Medium: 4-6 cores **≥2.3 GHz**. Large (100/100/10): 12–16 cores **≥2.4 GHz** (physical preferred), 32 GB RAM, 2–4 TB NVMe for archive, 1 Gbit/s.
- 4
Why 100 cameras is a lot of CPU and disk
For each motion camera there is a separate ffmpeg. On “normal” cores (≥2.4 GHz) 100 cameras ≈ 4–8 cores; at slow 1-1.5 GHz - multiply the estimate by 2-3. VMD on Hikvision (hybrid), not OpenCV for all 100 threads.
- 5
Video archive disc (rating)
With moderate motion activity: about 3–8 GB per day for 100 cameras (substream, debounce 30 s). For 30 days - 90–240 GB of clips; in active production ×2–4. Set the quota and cyclic re-recording in “Video archive → settings”.
- 6
Net
100 substreams of 0.5–1 Mbit/s - 50–100 Mbit/s constantly per server. Plus splashes when viewing the grid and TV Wall. Cameras and server are on the same L2 network with a reserve of bandwidth; Don't route RTSP over a narrow WAN.
- 7
Post-implementation check
Keep track of CPU, RAM and swap (admin → Server and services). SSH command: pgrep -c ffmpeg - the number of processes is close to the number of cameras with recording. Constant CPU >85% or active swap - increase resources or reduce load (VMD on cameras, fewer motion cameras, sub-streaming).