Flagship AI
Performance on a
Single Board

Powered by the Intel Core Ultra X7 358H, LattePanda Sigma 2 pairs heterogeneous computing with 64GB of LPDDR5X memory to deliver up to 180 TOPS of overall peak AI performance. It provides the speed and capacity for local AI agents and demanding workloads—without relying on cloud inference.

Ample Unified Memory
for Local AI Inference

• 64GB LPDDR5X-8533 RAM

  136.5 GB/s bandwidth for high-throughput MoE inference

• Up to 59GB Allocatable VRAM

  Enough room for a 35B-class quantized MoE model and the KV cache required by long-context AI agents

Works With Your AI Framework

Build and deploy with the tools you already know.

Local AI, Ready Out of the Box

With Ubuntu OS, NPU and GPU drivers, OVMS (OpenVINO Model Server), OpenClaw, and local AI models pre-installed on the built-in SSD, LattePanda Sigma 2 is ready for local AI—no driver setup, framework configuration, model downloads or cloud AI API required.
You can also install other AI agents and connect them directly to its local inference API.

[START]

[OK]

[OK]

[OK]

INFERENCE ENGINE

RUNNING

Accelerator Intel iGPU

Model Qwen3.6-35B-A3B

LOCAL INFERENCE API

ONLINE

Endpoint

http://localhost:8080/v3

LattePanda Sigma 2 — The Embeddable AI Single Board Server

The LattePanda Sigma 2 is a high-performance x86 AI single board computer (SBC) and embeddable AI server powered by the Intel Core Ultra X7 358H processor. Combining a 16-core CPU, Intel Arc B390 GPU with 12 Xe-cores, and an integrated Intel NPU, it delivers up to 180 TOPS of overall peak AI performance (INT8). With up to 64GB LPDDR5X-8533 unified memory, 136.5 GB/s memory bandwidth, and up to 59GB of allocatable VRAM, Sigma 2 is built for local AI inference, large language models (LLMs),Vision Language Models(VLMs), mixture-of-experts (MoE) models, AI agents, and long-context workloads without relying on cloud AI APIs.

Designed for edge AI computing and embedded AI applications, Sigma 2 supports Windows 11 and Ubuntu 24.04 LTS or later. Its ready-to-use configuration includes Ubuntu, GPU and NPU drivers, OpenVINO Model Server (OVMS), OpenClaw, and local AI models pre-installed on the built-in SSD, simplifying deployment of private AI agents and retrieval-augmented generation (RAG). Extensive connectivity includes dual 10GbE Ethernet, two Thunderbolt 4 ports, HDMI 2.1, USB 3.2 Gen 2, three M.2 M-Key expansion slots for NVMe SSDs and compatible AI accelerators, and OCuLink PCIe 4.0 x2 for external graphics card expansion. The onboard RP2350 MCU and GPIO support I2C, SPI, UART and ADC for sensor integration and actuator control.

With a compact 146 × 102 mm form factor and approximately 4.5W idle power consumption, LattePanda Sigma 2 is ideal for 24/7 local AI services, private AI servers, AI homelabs, NAS and virtualization, industrial edge computing, professional data acquisition, computer vision, and mobile autonomous robots. It combines high-performance AI computing, flexible PCIe expansion, and real-time hardware control in a single board for developers, makers, researchers, and system integrators building private local AI systems.