AE-3588LBT: Fanless Industrial Edge AI Computer | RK3588 32 TOPS

Choose the AE-3588LBT when you need RK3588 inference with room to grow: a 6 TOPS onboard NPU plus an M.2 slot for an additional NPU module, HDMI 4K input, 8K output and a -40°C to +80°C operating range in a compact cabinet-mounted box. The AE-3588NBT is the lower-cost indoor gateway with a fixed 6 TOPS NPU. The AE-3588BT adds RK3588J, 4000V isolated I/O and up to 58 TOPS for sites with long cables, high EMI or separate ground domains. See the AE-3588BT vs AE-3588LBT comparison.

It is the combined AI headroom of two engines: the 6 TOPS NPU built into the RK3588 and an M.2 NPU accelerator of up to 26 TOPS. The two engines do not pool into one 32 TOPS budget for a single model. Each runs its own models through its own SDK, so a typical design splits the pipeline, for example detection on the M.2 accelerator and classification or OCR on the onboard NPU.

No. The single M.2 M-Key slot takes either an NVMe SSD or one NPU accelerator. With the accelerator fitted, storage runs from the 128 GB eMMC (256 GB optional) or USB 3.2. If the project needs both NVMe storage and extra AI acceleration, look at the AE-3588BT.

Export the model to ONNX or TFLite, then compile it to the .rknn format with Rockchip RKNN-Toolkit2 for the onboard NPU. INT8 post-training quantization with a representative calibration set gives the best throughput. Models for the M.2 accelerator use that accelerator’s own toolchain. Accuracy after quantization depends on the model and calibration data, so validate it on your own production images.

Yes. It has one HDMI 2.0 input for 4K@60Hz capture and one HDMI 2.1 output up to 8K@60Hz, plus eDP, LVDS and MIPI DSI display options. A typical AOI or inspection pipeline captures from a vision source through V4L2, runs inference on the NPU and renders the overlay on the output. End-to-end latency depends on the source, model and rendering path, so measure it in your own setup.

No. The 2× RS-232/485 ports and 4-bit GPIO share a ground with the power input, which is the intended design for cabinet installations with a single-point equipotential ground. Route signal cables away from VFDs, relays and AC wiring, and add external isolators for long runs or ground-loop-prone devices. For sites without a unified ground or that need 4000V isolation, use the AE-3588BT. See industrial PC galvanic isolation.

The extruded aluminum unibody acts as the heatsink, and the unit is rated for -40°C to +80°C operation. Typical machine-vision duty cycles run without throttling. For sustained maximum CPU and NPU load near the top of the range, mount it in a ventilated enclosure or add airflow, and validate thermal behavior with your real workload.

The AE-3588LBT supports Android 12 with full root access, Debian 11 and Ubuntu 20.04. BITECH provides an open BSP based on Linux 5.10 / 6.1 LTS with kernel source, cross-compilation toolchains and defconfigs, so you can build out-of-tree modules, patch V4L2 for industrial cameras or deliver a custom production image.

It provides 2× Gigabit LAN, 2× RS-232/485, 4-bit GPIO, 2× USB 3.2 Gen1, 2× USB 2.0, HDMI 2.0 in, HDMI 2.1 out, eDP, LVDS and MIPI DSI. Wireless options include 4G/5G with GPS/BeiDou through Mini PCIe, optional Wi-Fi 6 and Bluetooth 5.2. The enclosure measures 148 × 126 × 40 mm.

Every AE-3588LBT runs an 18-hour full-load burn-in before shipment, and the platform includes hardware and software watchdogs and an RTC with backup battery for unattended operation. BITECH provides a controlled BOM with PCN notice, a 24-month warranty and OEM options for OS images, I/O layout and labeling. See how we test every unit and lifecycle management.