Inference where the work happens.

Same goal, two questions

Physical AI and Edge AI, working together.

Physical AI

What it does

Perceiving, reasoning, and acting in the real world through robots, vehicles, and drones, in a continuous sense-decide-act loop.

Edge AI

Where it runs

Locally on the device, not the cloud, for millisecond latency, privacy, and resilience when the link drops.

Safe physical action needs millisecond decisions. The cloud still trains the models; the edge runs them, on-device, even when the link drops.

A TOPS ladder

From a one-watt vision module to a 100-TOPS box.

AMD Ryzen Embedded 8000

Bedrock R8000

A sub-1-litre fanless box that stacks Ryzen AI with up to three Hailo accelerators past 100 TOPS.

CPU

8 cores / 16 threads · to 5.1 GHz

AI

100+ TOPS · Ryzen AI + Hailo

Memory

96 GB DDR5 ECC

I/O

USB4 40 Gbps · WiFi 6E · 5G

Hailo-15 vision SoC

Hailo-15 SOM

Up to 20 TOPS of vision inference on a 47 × 30 mm module sipping around a watt.

NPU

Up to 20 TOPS

Imaging

4K ISP · 2 × MIPI-CSI2

Memory

8 GB LPDDR4 · to 256 GB eMMC

Size

47 × 30 mm · ~1 W

NXP i.MX 8M Plus

i.MX8M Plus SOM

A 2.3-TOPS industrial module with TSN networking and a real-time M7, guaranteed to 2036. From $61.

CPU

4 × A53 + Cortex-M7

NPU

2.3 TOPS

Networking

2 × GbE (1 × TSN)

Longevity

To 2036 · from $61