Step 1 · Model

Start from the model.

Drop an ONNX file. We read it, nothing to fill in.

What we read from the file
File—
Takes—
Returns—
Parameters—
Compute—
Layers—
sha256—
read from the file · nothing is run · the file is not kept
Step 2 · Overview

The model, at a glance.

What the model is, how big it gets, and which boards can hold it.

Architecture
Edge boards · our catalogue
BoardFP32FP16INT8INT4Fits, and howLab
read from the file · memory from vendor specs · nothing is measured on this page
Step 3 · Boards

Which board for this model?

Say what matters to you; the ten board families in our lab are ranked as you move the sliders. Pick the ones to analyse layer by layer.

What matters to you · 0 to 10
Best match
BoardBoard priceAcceleratorMemoryHolds this modelIn our lab · 71 modelsPowerMatch
board prices checked at sellers on 2026-11-15 · specs from our catalogue · power as the vendors state it · speed estimated · lab columns measured on our boards
Step 4 · Layers

Where each layer would run.

On each board in our lab: on the accelerator, back on the CPU, or fused into its neighbour. Predicted from the file, by models trained on what we measured.

BoardLayers, in file orderAccel. · CPU · fusedCompute on accel.Held-out
predicted, not measured · whether the model compiles at all is not predicted · the file is not kept