AMD Kria AI Robotics Developer Platform — SOM, Carrier and Thermal Enclosure

Specifications

Overview

The AMD Kria AI Robotics Developer Platform is a complete, integrated robotics development kit consisting of a Kria AI SOM, a Spartan™ UltraScale+™ FPGA carrier card and a thermal enclosure. It is the reference platform for AMD’s physical AI push, announced at Advancing AI 2026 alongside the Ryzen AI Embedded X100 processor family.

The compute module carries an X199 processor — 16 Zen 5 CPU cores, an RDNA 3.5 GPU at 60 FP16 TFLOPS and an XDNA 2 NPU up to 50 TOPS — backed by 64 GB or 128 GB of LPDDR5X unified memory shared across CPU, GPU and NPU. The carrier’s Spartan UltraScale+ FPGA handles deterministic real-time I/O, sensor fusion and safety functions, which is why the platform can close control loops in 1.5 microseconds rather than being limited to perception workloads.

The I/O mix is explicitly robotics-shaped rather than general-purpose. Cameras attach through two FAKRA x4 connectors covering GMSL2/3 and GMSL1/2, an onboard IMU provides inertial sensing without an add-on board, and networking spans dual 5 GbE, dual 1 GbE with EtherCAT/TSN for servo and drive integration, and a 10 GbE QSFP uplink. Two CAN-FD buses and RS485 cover vehicle and industrial peripherals, with an A2B audio bus included for in-cabin or machine-audio use.

Storage and expansion are handled by an M.2 2280 Key M NVMe slot and an M.2 2230 Key E socket for WiFi/Bluetooth, plus an OcuLink PCIe connector and a Pmod interface for rapid prototyping. AMD publishes open-source baseboard schematics so teams can move from the kit to a production carrier without reverse-engineering the reference design, and the whole stack runs AMD’s open robotics SDK on ROCm.

AMD claims 3.4× the real-time response performance of the NVIDIA Jetson Thor platform based on its commissioned OpenNav Robotics Workload Benchmark. The kit was sampling with early-access customers at announcement with general availability planned for Q4 2026, and CNX Software reports indicative pricing expected to exceed $5,000 — deliberately positioned against the NVIDIA Jetson AGX Thor Developer Kit. QS Compute can supply the platform along with the Kria AI SOM for production builds.

Key Benefits

Everything in one box — SOM, FPGA carrier and thermal enclosure. GMSL2/3 and GMSL1/2 camera inputs on FAKRA connectors for automotive-grade vision. EtherCAT/TSN networking integrates directly with servo drives and real-time machine networks. Onboard IMU removes an add-on sensor board. 128 GB unified LPDDR5X supports larger perception models without memory staging. Open-source baseboard schematics de-risk the move to a production carrier.

Applications

Autonomous mobile robots and mobile manipulators; humanoid and legged robot development; factory automation with real-time servo integration; agricultural and off-highway autonomy; multi-camera perception and sensor fusion research; safety-rated robotics prototyping; physical AI and embodied intelligence research programmes.

Related Products