Section 3 — Latency at the Edge
07 / 12
Inference benchmark
The model is fast. The network is slow.
P99 Latency (ms)
0
10
20
30
40
50
12 ms
Cirrus PoP
private
18 ms
Regional Edge
hybrid
28 ms
Self-hosted
on-prem
34 ms
Hyperscaler
us-west-2
41 ms
Hyperscaler
eu-west-1
54 ms
Multi-tenant
shared API
−79%