What problem does it solve? Choosing the wrong RDK board or over-estimating what a board can run (e.g. expecting an X5 to smoothly run a 7B LLM) leads to bad purchasing decisions and failed projects. This Skill grounds board selection, model-feasibility judgments, and cross-platform comparisons in official D-Robotics benchmark data instead of guesswork. ## Core Features & Use Cases - Board selection cheat-sheet: Compares X3, X5, Ultra, S100, S100P, and S600 by compute, CPU/MCU, RAM, and model artifact format (.bin vs .hbm), with one-line picks per use case. - LLM/VLM expectation calibration: Quotes official decode TPS benchmarks (e.g. S100P 7B q8 ≈ 6.7 TPS, S600 Qwen3-8B w4 ≈ 31.4 TPS) to distinguish "can run" from "usable", and explains the three LLM routes (hobot_llamacpp, hobot_llm, native Ollama CPU-only). - Cross-platform comparison: Positions RDK honestly against Jetson Orin Nano, Raspberry Pi 5 + AI HAT+, and RK3588 boards, noting different TOPS units and each platform's strengths. - Use Case: A user asks "RDK X5 能跑 DeepSeek 吗?" — the Skill asks which DeepSeek size, explains X5 handles ≤2B quantized models only, warns that native Ollama is CPU-only, and routes the actual build to rdk-llm-deployment. ## Quick Start Ask which RDK board you should buy for your project, or whether a specific model like YOLO, DeepSeek, or InternVL can run on your board.
