Edge AI Simulator | mysimulator.uk

Edge AI Network

Edge devices 12

Local inference share 60%

Model compression 0.50×

Network latency 160 ms

Edge avg latency

–

Cloud avg latency

–

Bandwidth saved by edge devices

–

How it works

Each device around the ring periodically generates an inference request. A device assigned to edge mode runs a compressed model locally — its indicator light pulses green while it "thinks," and the request never leaves the device. A device assigned to cloud mode instead packages the request and ships it up to the central server: the packet must travel the round-trip network distance, wait for the server to run the full-size model, and travel back before the result lands.

Model Size = Original Model Size × Compression Factor
Edge latency  ≈ 40ms + Compression × 260ms
Cloud latency ≈ Network round trip + fixed server compute

Edge devices — how many nodes sit on the ring; more devices means more concurrent requests to compare.
Local inference share — what fraction of devices run compressed models on-device instead of calling the cloud.
Model compression — a smaller factor means a smaller, faster (but less capable) on-device model; a larger factor keeps more accuracy at the cost of speed.
Network latency — the round-trip time to the cloud server; edge inference is immune to it entirely, which is the whole point of moving intelligence to the edge.

Real-world relevance: this is the same trade-off behind on-device voice assistants, smart cameras and industrial sensors — trading some model accuracy for latency that no network connection can beat, while also cutting the bandwidth every cloud round trip would have consumed.

Edge device — local inference Cloud device — request packet Cloud server

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.