AI / ML Infra · Benchmarking

Aether

Edge vs. cloud ML inference benchmarking. Aether runs Transformers.js v3 client-side via WebAssembly and WebGPU head-to-head against a persistent Node.js server backend, and pinpoints exactly where local inference beats the network round-trip — and where it doesn't.

Live Demo
What it does

Find the tipping point between edge and cloud.

⚙️

One-click automated suite

Runs all 9 testing matrices sequentially — short, medium, and long text payloads against client WASM, client WebGPU, and server targets — and generates a grouped crossover chart to isolate the tipping point.

📊

Targeted diagnostics

Manually override the payload and target to break down RTT latency, pure inference compute, transit/worker overhead, and cold-start graph compilation for any single run.

🖥️

Live execution console

Streams execution logs and live Railway network latency as each benchmark runs, so you can watch WASM, WebGPU, and server inference compete in real time.

Tech stack
Transformers.js v3 WebGPU WebAssembly Node.js Deployed on Railway
Live preview

Try it here.

edge-vs-cloud-inference-benchmarks-production.up.railway.app