How to choose: answering the main questions
CPU or GPU for analytics?
For up to 8–12 channels of motion and line-crossing detection, the CPU is sufficient. Facial, license plate, and PPE recognition on 16+ channels requires a GPU: a single L4/RTX A-series card replaces dozens of cores and removes scaling limits.
Can we combine recording and analytics on a single server?
Yes, that is standard: the disk array handles the archive, while the GPU processes neural networks. It is crucial to separate the workloads—use dedicated volumes for the archive and separate resources for analytics; otherwise, peak write operations will throttle the recognition process.
What about VMS compatibility?
Macroscop, Trassir, Axxon Next, Milestone — each has its own requirements for GPU and drivers. We build servers for specific VMS and channel counts, with headroom for adding analytics modules.