High CPU or GPU load and missed counts

Signs the PC is overloaded

• Task Manager shows the processor near 100% for long periods.
• The live picture in the app stutters, tracking boxes lag behind people, or fast objects are missed.
• Counts fall at busy times, exactly when you need them.

Lighten the load

1. 1

   Lower the working resolution

   In each camera’s settings, reduce the resolution (for example to 360p). Counting rarely needs more; faces are the exception.

2. 2

   Turn on hardware decoding

   If the PC has a graphics card, tick hardware decoding for each camera to take video decoding off the processor.

3. 3

   Narrow the area of interest

   Analyse only the part of the picture where the counting happens.

4. 4

   Use basic motion where it is enough

   A simple in-and-out count at a quiet door does not need an AI profile, and basic motion uses about a tenth of the processing.

5. 5

   Stop what you do not need

   Set cameras you only view or record to analytics off, switch off local recording you do not use, and keep other software off the PC.

When to add a computer

As a guide for 1080p streams:

• Intel i9, 64 GB RAM, GeForce RTX: 30+ feeds on the AI profiles, 40–80 feeds on basic motion
• Intel i7, 32 GB RAM: 7–15 feeds on the AI profiles, 25–35 feeds on basic motion
• Intel i5, 16 GB RAM: 5–10 feeds on the AI profiles, 17–25 feeds on basic motion
• Intel i3, 8 GB RAM: 3–6 feeds on the AI profiles, 10–20 feeds on basic motion
• Intel Celeron / Atom, 4 GB RAM: Not supported on the AI profiles, 3–7 feeds on basic motion

If you are at the top of the band for your machine, split the cameras across two PCs. Each becomes its own location, and reports can combine them.