Knowledge base

Technical answers, sizing guidance and deployment notes from the NetBytes engineering team.

GPU and AI Systems

Choosing between 2U, 3U and 4U for a GPU server Compare 2U, 3U and 4U GPU servers by qualified accelerator capacity, drive layout, cooling geometry, chassis depth and actual rack power limits.

Thin client or compact PC for a VDI rollout Choose a thin client for controlled, network-dependent VDI seats. Choose a compact PC where peripherals, offline work, graphics or local applications require dependable endpoint compute. The correct comparison includes management effort, VDI infrastructure and support costs, not only device price.
RAID 6 or RAID 5 for Surveillance Recording For any surveillance array built on drives 8TB and above, RAID 6 is the correct default, not because it is more capacity-efficient, but because RAID 5's rebuild window on large drives has become a genuine data-loss risk. Here is the math that decides it.

Tower server or rack server for a first server room Choose the form factor by checking the rack, equipment location, expected growth and maintenance access first. CPU core count usually comes later because tower and rack platforms overlap considerably.

How to Size Recording Storage for a 128-Camera Deployment The arithmetic behind sizing storage for a 128-camera site is not complicated, but it has enough moving parts — bitrate, retention, RAID efficiency, headroom — that skipping one step is how installers end up 25% short on delivery day. Here is the working, step by step.

How Many Virtual Machines Fit on One Host There is no fixed VM count for a host. Calculate CPU capacity from reserved physical cores and an explicit vCPU ratio, calculate memory after reserves and headroom, then use the lowest result after checking storage, network, licensing and N+1 failover capacity.

When a workstation actually needs ECC memory ECC memory is justified when a memory error could invalidate an expensive simulation, corrupt an important output or leave an engineer unable to trust the result. Runtime matters, but so do memory footprint, checkpointing, validation and rerun cost.