AE-760EBT: Intel® Core™ i9 + NVIDIA® RTX MXM Industrial Edge AI PC

The mainboard hard-caps the MXM 3.1 Type-B slot at 115 W Total Graphics Power.

This is a deliberate design decision, not a limitation we discovered late. A compact industrial chassis cannot dissipate 175 W of GPU heat continuously while also sustaining i9 turbo. Capping in the mainboard means performance is deterministic and repeatable instead of throttling unpredictably under sustained load.

What this means per module:

• RTX 4060 MXM (115 W) — runs unthrottled at full factory clocks. This is the reference SKU behind all published benchmarks.
• RTX 4070 / 4090 / 5080 / 5090 MXM — can be fitted, but operate at the 115 W tier. You gain VRAM capacity and Tensor Core count, not raw clock headroom.

If your workload requires sustained 175 W GPU power, this chassis is not the right fit — and we will say so at RFQ rather than after delivery.

The BIOS applies power profiles tailored for industrial deployment: PL1 (long-term) locked at 125 W, PL2 (short-term) up to 219 W for 28 seconds.

To sustain this, the system uses 4× composite copper heat pipes mated directly to the CPU die, drawing heat to an aluminium fin array cooled by twin 80 mm high-static-pressure PWM fans.

In a 55°C ambient environment running continuous validation workloads, CPU core temperature stabilises below 92°C, allowing P-cores to hold a steady 3.0–3.5 GHz baseline indefinitely. No thermal throttling was observed up to 320 W combined system load.

Completely independent. There is no single-channel switch chip bridging traffic.

The mainboard integrates 4× independent Intel i226-IT 2.5GbE controllers, each on a dedicated PCIe lane.

Aggregating four GigE Vision cameras at full 2.5 Gbps line speed puts total system throughput near 10 Gbps. This native architecture prevents internal bus congestion, packet collision and dropped frames at the edge — the failure mode that makes bridged four-port designs unusable for multi-camera inspection.

Chassis dimensions are 240 × 200 × 60 mm] at approximately 6.0 kg, with dual 80 mm fan exhaust on one face.

Airflow clearance: allow at least 10 mm of unobstructed space in front of the fan exhaust and intake faces. Mounting the unit flush against a cabinet wall on the exhaust side will recirculate hot air and reduce the sustained power envelope.

Mounting options: wall, side, DIN-rail and VESA. The fan brackets sit outside the sealed internal chassis, so field replacement does not require opening the thermal assembly.

Storage configuration: 1× M.2 2280 PCIe NVMe + 1× 2.5″ SATA SSD/HDD + 1× mSATA.

Why this matters for edge LLM workloads: a 34 B model quantised to Q4 occupies roughly 20 GB on disk. Load time differs substantially by interface — SATA at ~550 MB/s takes well over 30 seconds, while PCIe NVMe loads the same model in a few seconds.

For deployments that load models on boot or swap between models at runtime, specify the NVMe configuration at RFQ. For deployments that load once and keep the model resident in VRAM, SATA is sufficient.

The AE-760EBT accepts a wide-range DC 9–36 V input with reverse-polarity protection, over-current cutoff, and transient voltage suppression (TVS) diodes on the input stage.

For the typical case — a 12 V or 24 V battery bus sagging toward 9 V for up to 100 ms during high-torque motor startup — the onboard DC-DC converters and solid-state capacitor bank hold internal rails stable, preventing resets or unplanned shutdowns.

Input is available on a 5.5×2.5 mm DC jack, a 3-pin Phoenix terminal, and a 2-pin remote power on/off terminal, so it drops into existing vehicle and cabinet wiring without an adapter harness.

Yes — both CAN FD ports carry 4000 V isolation withstand to break ground loops and block high-voltage spikes from switching contactors and heavy industrial equipment. The four RS-232/485 ports and the 8 DI / 8 DO block are opto-isolated as well.

For impedance matching on long field-bus runs, the mainboard provides onboard 120Ω termination resistors on jumper pins. Two terminators at the two ends of the bus read approximately 60Ω across the pair — the standard check. This removes the need for external termination dongles and eliminates reflection-induced data loss.

Because the AE-760EBT uses dual 80 mm exhaust fans rather than passive conduction, the enclosure must move air. Requirements:

• Filtered air intake paths or louvers in the cabinet — a fully sealed, unventilated enclosure will not work with this platform
• Operating range −20°C to +60°C at the intake, storage −40°C to +80°C
• Clearance in front of both the intake and exhaust faces so hot air is not recirculated

For sealed, dust-prone or fully passive enclosures, an ARM-based platform without a discrete GPU is the better architectural fit — the trade-off being that you lose the full CUDA and Windows software stack.

It is a discrete hardware security chip on the Infineon SLB9670 platform, not CPU-simulated firmware TPM (fTPM), which avoids the class of vulnerabilities associated with fTPM implementations.

For grid, critical infrastructure and high-security industrial deployments, BITECH supports factory-level UEFI Secure Boot customisation. Your PK, KEK and db keys can be flashed into custom BIOS firmware during production, locking the boot state against physical firmware injection.

Two different clocks, and it is worth being precise about which is which.

The x86 mainboard, chassis and I/O are built on the Intel Embedded Roadmap under a BOM freeze policy once an OEM configuration is approved, with 10-year availability and formal PCN issued 6 to 12 months ahead of any unavoidable component change.

MXM GPU modules follow NVIDIA's own generational cycle, which is shorter. We track module availability and issue PCN with a last-time-buy window when a module reaches EOL. Because MXM 3.1 Type-B is a standard interface, a successor module can be qualified into the same chassis without redesigning your enclosure or re-certifying the mechanical envelope.

Warranty is 24 months on the complete unit. Every unit passes 18-hour PassMark BurnInTest under full CPU and GPU load before shipment.