LattePanda Mu Ultra Release Image © LattePandaLattePanda Mu Ultra Release (Image © LattePanda)

The module integrates the CPU, GPU, and NPU of the Intel Core Ultra series, achieving a total AI performance of 115 TOPS (INT8), with the dedicated NPU contributing 47 TOPS. By executing AI workloads locally, the system reduces latency, improves data privacy, and minimizes reliance on cloud infrastructure.

Technical tests using OpenVINO GenAI and INT4 quantization on the iGPU show text generation speeds of 55 tokens per second with Qwen3.5-2B and 18 tokens per second with Qwen3.5-9B. These figures demonstrate the module’s ability to handle responsive, local, dialogue-based AI for offline document processing and the retrieval of private knowledge bases.

To support demanding AI workloads, the LattePanda Mu Ultra is equipped with 16 GB of LPDDR5X-8533 memory. The system allows up to 11.6 GB of this memory to be allocated as video memory (VRAM), providing the necessary bandwidth for large language models (LLMs) and real-time sensor processing. Additionally, the module has an idle power consumption of 2.5 W, making it suitable for always-on and portable applications.

LattePanda Mu Ultra LLM LeistungLattePanda Mu Ultra LLM Leistung (Image © LattePanda)

The module is compatible with Windows and Linux, ensuring integration into the existing x86 software ecosystem. It supports industry-standard AI frameworks and tools such as Intel OpenVINO, llama.cpp, and Ollama, allowing developers to implement AI capabilities without having to redesign existing software stacks.

The LattePanda Mu Ultra uses a standardized form factor that is compatible with the Mu family of carrier boards, such as the 3.5-inch Lite, Mini, and M.2 M-Key boards. This modularity enables system upgrades without requiring new carrier board designs. Connectivity options include PCIe 4.0, USB 3.2 Gen 2, USB 2.0, HDMI, DisplayPort, eDP, GPIO, UART, and I2C.

Main Application Areas

The module is designed for five key application areas:

  • AI terminals: Local execution of LLMs for translation devices and voice assistants.
  • Portable measuring devices: AI-powered real-time analysis for spectrum analyzers and diagnostic tools.
  • Autonomous mobile robots (AMRs): Processing for SLAM, sensor fusion, and autonomous navigation.
  • Service robots: Integrated processing for multimodal AI with image and speech recognition.
  • Edge image processing: Low-latency defect detection and high-resolution video processing for smart cameras.

The LattePanda Mu Ultra is currently available in the official online store for €548.