Up to 72 TOPS of NPU capacity on Raspberry Pi 5
In a publication dated August 11, 2025, Sipeed introduced the Maix4-HAT, with an AXera AX650N NPU, up to 72 TOPS in INT4 or 18 TOPS in INT8, and 8 GB of LPDDR4x RAM for edge Transformer inference.
Source: Sipeed Maix4-HAT leva NPU AX650N ao Raspberry Pi (cnx-software.com). Text prepared with AI from this source.
What happened and what to do
In a publication dated August 11, 2025, Sipeed introduced the Maix4-HAT for Raspberry Pi 5. The board integrates an AXera AX650N NPU, rated at up to 72 TOPS in INT4 or 18 TOPS in INT8, alongside 8 GB of LPDDR4x RAM. It is intended to evaluate inference of quantized Transformer models on edge AI devices.
A company could respond with a local inference pilot: select models and workloads, measure latency, memory use, and accuracy, and define how to update models and monitor devices. The consultancy can assess whether this class of hardware fits the use case and design integration with the operation’s systems and data.
How the consultancy can help
Wendelmaques can diagnose the opportunity for edge inference, define evaluation criteria, and deliver a scoped implementation covering integration, monitoring, and an operating plan.
Next step
Send a short description of your use case to receive a scoped proposal for diagnosis and implementation.
Let's assess your project
I reply with scope and a proposal.