What this checklist covers
Use the WF-RK3588-E0 reference to define an edge AI board project around a measurable workload, supported software and the actual enclosure. This checklist describes evaluation work; it is not a customer deployment report, benchmark result or guarantee of a particular inference rate.
Define the workload before comparing compute claims
Record the model architecture, input resolution, precision, batch size, concurrent streams, latency target and accuracy requirement. Confirm toolchain and operator compatibility on the exact BSP. Peak processor figures do not establish application throughput.
Interfaces and sustained operation
List cameras, displays, storage, network and serial devices. Test concurrency and recovery using the actual peripherals. Measure sustained behavior in the intended enclosure and power configuration; distinguish a thermal limit from a short demonstration.
Make acceptance reproducible
- Identify the PCB revision, memory configuration, firmware and model checksum.
- Keep input data, measurement method and test duration with the results.
- Test cold boot, interrupted network, storage pressure and update recovery.
- Define pass/fail limits agreed for the application rather than copying generic factory claims.
Documents and responsibility
Request the current model specification, pinout, BSP release notes, known limitations and applicable compliance documents. Confirm software maintenance, change notification, sample scope and delivery terms in the project agreement.
RK3588 project planning · Quality and test scope · Discuss the workload
