From Sensor to GPU on a Single Vendor’s Hardware
Physical AI systems rarely run on one processor. A real-time controller handles sensors, safety and actuation, and a GPU handles perception and planning. Connecting the two is where integration time goes, especially when the hardware comes from different vendors.
SolidRun builds both NXP-based edge modules and NVIDIA-based AI systems, and we are the first to demonstrate a complete, self-contained i.MX 95 to NVIDIA Holoscan to GPU implementation on hardware from a single vendor. The SoM captures camera and sensor data, processes it on its integrated ISP and NPU, and streams it as timestamped, low-latency Holoscan data over 10GbE to a SolidRun NVIDIA-based AI system. The same link carries commands back to the SoM’s real-time Cortex-M7 core, so one cable closes the loop between sensing and action.
Because the implementation is software-defined, it handles more complex workloads than fixed-function front ends. And because both ends of the link come from SolidRun, you have one hardware partner for the whole pipeline.