Free tool
Size your Edge deployment
How many GPUs, how much bandwidth, how much per month — for a distributed vision deployment (occupancy, counting, crowd, demography, presence). Pure client-side math, no account needed.
Where these numbers come from
Tesla T4 per-model throughput and VRAM were benchmarked on a real g4dn.xlarge (FP16, one model per process — matching the edge supervisor). Other GPUs are scaled from that baseline by relative FP16 throughput with a conservative efficiency factor: treat them as ±30%. Cloud prices are approximate on-demand list prices circa 2026 — they drift by region and commitment, so verify with the provider before quoting a customer.
Need help sizing a real deployment?
We run the numbers with you and design the cluster — consulting, not a per-camera sale.