MYiR’s new MYC-JR3576 module pairs a Rockchip processor with a 6-TOPS NPU, and is shipping now from $192. Its standard SMARC form factor gives industrial and embedded developers another concrete option for building edge AI systems, without needing to design around a custom board from scratch.
Watch Desk analysis
What happened
The MYC-JR3576 is a system-on-module using the Rockchip RK3576, or the industrial-grade RK3576J. Its NPU supports INT4, INT8, INT16, BF16 and TF32 operations, and the module offers up to 8GB of LPDDR5 memory and 64GB of eMMC storage. MYiR also offers a compatible development board.
CNX Software says the modules and kits are shipping now. Listed prices are $209 for an 8GB/64GB commercial-grade module, $192 for a 4GB/32GB industrial-grade module, and $279 or $293 for a development kit, depending on the selected module. MYiR’s Linux 6.1 BSP and Android 15 support are available to customers, not publicly.
Why it matters
The 6-TOPS NPU and industrial temperature option make this more than a general-purpose embedded board with “AI” sprinkled over the spec sheet. Developers can choose between commercial and industrial versions, with the latter rated for temperatures from -40°C to 85°C, and connect the module through a standard 314-pin MXM interface.
That does not tell us how quickly any particular model will run, or which workloads will fit within its memory. But the listed price, form factor and software support give developers useful starting points for assessing whether it belongs on a shortlist.
Our read
This is a practical edge-AI building block, not a magic model-in-a-box. The standard module format and published pricing make it easier to evaluate than a mystery board with a dramatic TOPS number and no shopping details. The software access restriction is worth factoring in before planning a project around it.
What to watch
- Whether MYiR publishes model-specific inference benchmarks and power figures.
- Which software packages and AI frameworks are available to customers.
- Whether the industrial-grade version is tested in complete systems across its rated temperature range.
Discussion spark: For edge-AI projects, would you prioritise a standard module with published pricing, or wait for independent benchmarks showing what its NPU can actually run?
Sources and evidence
- Source update (27 October 2026, 00:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.