AMD Kria AI SOM — COM‑HPC Module with Ryzen AI Embedded X100 and Integrated FPGA

Specifications

Overview

The AMD Kria AI SOM is a COM-HPC system-on-module that combines a general-purpose CPU, a GPU, an NPU and an FPGA on a single module. Announced at AMD’s Advancing AI 2026 event, it is built on the Ryzen AI Embedded X100 series processor — specifically the X199 — with 16 Zen 5 CPU cores, an RDNA 3.5 GPU rated at 60 FP16 TFLOPS, and an XDNA 2 NPU delivering up to 50 TOPS.

The architectural bet is unified memory. CPU, GPU and NPU all address the same pool of up to 128 GB of LPDDR5X rather than staging data across separate memory spaces, which removes the copy overhead that dominates perception and sensor-fusion pipelines on architectures with discrete memory. The integrated FPGA fabric handles real-time I/O, sensor fusion and safety functions deterministically, and AMD specifies control-loop closure at 1.5 microseconds — a figure that matters for motor control and safety-rated robotics, not just throughput benchmarks.

AMD also made an explicit standards choice: the module uses the open COM-HPC form factor rather than AMD’s historical proprietary module standard, and open-source baseboard schematics are published to accelerate production designs. The software stack is similarly positioned as open and non-vendor-locked, built on ROCm with optimised AI frameworks. For teams that have been constrained by a single-vendor robotics compute ecosystem, that combination is the product’s differentiator.

Industrial qualification is addressed directly. AMD states the Kria/X100 line carries extended reliability testing, burn-in, expanded test coverage and the fit rates required for industrial application — a deliberate contrast with consumer Ryzen parts. Module availability from ODM partners is planned for Q4 2026, with modules from Arbor, congatec, iBase, IEI, Sapphire and Seavo in development, and the X100 series committed to availability at least through 2037.

AMD claims 3.4× the real-time response performance of NVIDIA’s Jetson Thor platform based on an AMD-commissioned OpenNav Robotics Workload Benchmark published in July 2026. QS Compute supplies the Kria AI SOM and companion Kria AI Robotics Developer Platform through distribution channels; contact us for module and carrier configuration.

Key Benefits

CPU, GPU, NPU and FPGA on one COM‑HPC module. Unified memory up to 128 GB LPDDR5X eliminates cross-engine data staging. 1.5 µs control-loop closure supports deterministic real-time and safety-rated robotics. Open COM-HPC standard with published baseboard schematics avoids vendor lock-in. Extended industrial qualification including burn-in and industrial fit rates. Availability through 2037 with six ODM partners building modules.

Applications

Autonomous mobile robots (AMR) and mobile manipulators; humanoid and legged robotics; factory robot perception and motion planning; real-time motor and safety control loops; multi-sensor fusion for industrial autonomy; machine vision with deterministic actuation; physical AI research and development platforms.

Related Products