LattePanda Sigma 2 is an embeddable AI single board server built for local AI workloads. Powered by the Intel Core Ultra X7 358H processor (Panther Lake) with 64GB of LPDDR5X memory, it delivers up to 180 TOPS of total AI performance across its CPU, GPU and NPU—enough to run 35B-class quantized MoE models and AI agents entirely on-device. With dual Thunderbolt 4 ports, dual 10Gb Ethernet ports, four M.2 slots, OCuLink, and rich I/O on a 3.5-inch board, it brings scalable AI power to local AI appliances, homelabs, and robots.
Flagship AI Performance on a Single Board
The Intel Core Ultra X7 358H brings a 16-core CPU, an Intel Arc B390 GPU with 12 Xe-cores, and an Intel NPU together on one board. These compute engines deliver up to 180 TOPS of combined peak INT8 AI performance: 122 TOPS from the GPU, 50 TOPS from the NPU, and 8 TOPS from the CPU.
From local model inference to application processing and graphics, this architecture provides the computing foundation for AI servers and embedded systems.
.jpg "LattePanda Sigma 2 - Flagship AI Performance on a Single Board")
Expand Local AI Compute
Extend beyond onboard compute with compatible M.2 AI accelerators and external GPUs. M.2 M-Key, OCuLink, and Thunderbolt 4 provide multiple expansion paths as local AI projects grow.
Additional compute devices provide resources for more models and agent services, supporting deployments that bring multiple local AI workloads together on one platform.
.jpg "LattePanda Sigma 2 - Flagship AI Performance on a Single Board")
.jpg "LattePanda Sigma 2 - Flagship AI Performance on a Single Board")
Unprecedented Leap in Local Compute
Run 35B-class quantized MoE models directly on the board and bring model inference closer to local applications and data. The local LLM benchmarks show both input-processing and text-generation speeds, providing a practical view of model performance.
General computing takes a substantial step forward as well. The 16-core CPU and Intel Arc B390 GPU deliver higher CPU and graphics benchmark scores than the previous-generation LattePanda Sigma, supporting demanding applications alongside local AI workloads.
.jpg "LattePanda Sigma 2 - Unprecedented Leap in Local Compute")
Ample Unified Memory for Local AI Inference
LattePanda Sigma 2 features 64GB of dual-channel LPDDR5X-8533 memory with 136.5GB/s of theoretical bandwidth. Large capacity provides room for model weights and runtime data, while high bandwidth supports the data-intensive demands of local inference.
Up to 59GB of this unified memory can be allocated for GPU use, providing space for 35B-class quantized MoE models and the KV cache used to retain conversation context. The result is a platform built for local agents working with longer conversations and substantial inputs.
.jpg "Lightning-Fast Unified Memory for Local AI Agents")
Compact by Design, Ready to Embed
Local AI compute, high-speed networking, and extensive I/O fit within a 3.5-inch board measuring 146 × 102mm. The compact footprint supports integration into local AI appliances, mobile robots, and instruments where internal space matters.
Compute and hardware connectivity stay together on one board, creating a compact foundation for embedded AI systems.
.jpg "LattePanda Sigma 2 - Compact by Design. Ready to Embed.")
Fast Under Load, Efficient at Idle
LattePanda Sigma 2 delivers nearly twice the text-generation speed of the previous-generation Sigma in the illustrated local AI test, while average DC input power increases from 35W to 38W.
At idle, average DC input power drops to 4.5W, compared with 5W for the previous generation. Higher efficiency during inference and lower idle power make it a practical foundation for local AI services that remain available around the clock.
.jpg "LattePanda Sigma 2 - Fast Under Load, Efficient at Idle")
Windows & Ubuntu Support
LattePanda Sigma 2 supports Windows 11 and Ubuntu 24.04 LTS or later, providing operating-system options for application development and deployment.
The included SSD comes with Ubuntu pre-installed, providing a starting point for local AI projects.
Support for Popular AI Software Frameworks
Build local AI applications with Intel OpenVINO, Ollama, and llama.cpp. OpenVINO provides acceleration options across the CPU, GPU, and NPU, while Ollama and llama.cpp offer additional workflows for local model deployment.
From language and vision-language models to YOLO object detection and speech processing, the platform supports a broad range of AI development tasks through compatible software tools.
.jpg "LattePanda Sigma 2 - Support for Popular AI Software Frameworks")
Local AI, Ready Out of the Box
The included 1TB NVMe SSD comes pre-installed with Ubuntu, GPU and NPU drivers, OpenVINO Model Server (OVMS), OpenClaw, and local AI models. Drivers, inference software, and models are already in place, reducing the setup work required to begin local AI development.
Start the local inference API to provide on-device model inference for OpenClaw and other compatible agents. The local endpoint also provides a foundation for custom applications, with no cloud inference API required.
.jpg "LattePanda Sigma 2 - Local AI, Ready Out of the Box")
Complete I/O, All on One Board
Dual Thunderbolt 4 ports, dual 10Gb Ethernet, USB 3.2 Gen 2, HDMI 2.1, eDP, serial interfaces, and GPIO connect local AI compute to networks, displays, peripherals, and physical hardware.
Three high-power M.2 M-Key slots accommodate NVMe SSDs and compatible AI accelerators, alongside an M.2 E-Key slot for additional expansion. OCuLink and Thunderbolt 4 provide connection options for external GPU docks, bringing storage, networking, and compute expansion together on one board.
.jpg "LattePanda Sigma 2 - Complete I/O. All on One Board")
.jpg "LattePanda Sigma 2 - Expand Local AI Compute")
One Board for AI and Hardware Control
The onboard RP2350 microcontroller brings programmable hardware control alongside Intel-powered AI computing. While the Intel processor handles model inference and application logic, the RP2350 handles sensor inputs and actuator-control tasks.
Preloaded MicroPython firmware provides a starting point for programming hardware interfaces, including I2C, SPI, UART, and analog inputs through GPIO. Together, the two processors form a foundation for robotics, interactive devices, and sensor-connected AI systems.
.jpg "LattePanda Sigma 2 - From Local AI to Real-World Action")
Professional Technical Support and Integration Help
The LattePanda team provides technical support for installation, configuration, troubleshooting, and system integration, helping projects move from initial setup toward deployment.
Technical enquiries: techsupport@lattepanda.com
Customization for Project Requirements
LattePanda Sigma 2 offers customization options for system integration, including port removal, custom boot screens, BIOS feature adjustments, and tailored operating systems.
Customization enquiries: solution@lattepanda.com
.jpg "LattePanda Sigma 2 - Customization for Project Requirements")
Video: LattePanda Sigma 2 - The Embeddable AI Single Board Server
Features
- Intel Core Ultra X7 358H processor with 16 cores, 16 threads, and a maximum frequency of 4.8GHz.
- Up to 180 TOPS of combined peak INT8 AI performance across the CPU, GPU, and NPU.
- Intel Arc B390 GPU with 12 Xe-cores and up to 122 TOPS of peak INT8 AI performance.
- 64GB dual-channel LPDDR5X-8533 unified memory, with up to 59GB allocatable for GPU use.
- Included 1TB NVMe SSD with Ubuntu and a pre-installed local AI software environment.
- Dual Thunderbolt 4, dual 10Gb Ethernet, three high-power M.2 M-Key slots, one M.2 E-Key slot, and OCuLink.
- Onboard RP2350 microcontroller with preloaded MicroPython firmware for sensor interfaces and hardware control.
- Compact 3.5-inch board format measuring 146 × 102mm.
- Support for Windows 11 and Ubuntu 24.04 LTS or later.
Applications
- Local AI model and agent servers
- Embedded AI appliances
- Homelab servers and development systems
- Robotics and sensor-connected AI projects
- Interactive devices with AI-driven hardware control
- Compact instruments with local AI processing
.jpg "Applications of LattePanda Sigma 2 - The Embeddable AI Single Board Server (Intel Core Ultra X7 358H, 64GB RAM, 1TB SSD)")
Specifications
- Processor: Intel Core Ultra X7 358H; 16 cores, 16 threads, up to 4.8GHz
- GPU: Intel Arc B390; 12 Xe-cores, up to 2.5GHz
- NPU: Intel NPU
- Co-processor: RP2350 MCU
- Memory: 64GB dual-channel LPDDR5X-8533
- Storage: 1TB M.2 NVMe SSD
Display
- 1 × HDMI 2.1
- 1 × eDP 1.5, 4-lane
- DisplayPort 2.1 via USB-C (Thunderbolt 4)
Expansion and Connectivity
- 2 × Thunderbolt 4, 40Gbps
- 1 × OCuLink, PCIe 4.0 ×2
- 2 × USB 3.2 Gen 2 Type-A, 10Gbps
- 2 × USB 2.0 Type-A, 480Mbps
- 1 × USB 2.0 header
- 2 × 10Gb Ethernet, RJ45
- 1 × 3.5mm microphone/headphone combo jack
- 1 × RS-232/485 header
- 1 × GPIO header
- 1 × front panel header
- 1 × front audio header
- 3 × M.2 M-Key: PCIe 5.0 ×2 / PCIe 4.0 ×4 / PCIe 4.0 ×1
- 1 × M.2 E-Key: PCIe 4.0 ×1, USB 2.0, CNVio
Power
- DC input: 15–20V
- USB-C PD input: 20V
Operating System Support
- Windows 11
- Ubuntu 24.04 LTS or later
Operating Environment
- Operating temperature: 0–40°C
- Relative humidity: 0–80%
Dimensions
- Board format: 3.5-inch
- Size: 146 × 102mm
Documents
Package contents
- LattePanda Sigma 2 Single Board Server (358H, 64GB) x1
- Active Cooling Fan (Assembled) x1
- M.2 NVMe SSD WD SN7100 1TB (Assembled, Pre-installed Ubuntu OS) x1
- Metal Protective Bottom Plate x1
- Thermal Pads for M.2 SSD x3
- Power Adapter with DC cable x1
- AC Cable(EU & US & UK) for power adapter x3
- RTC Battery(Assembled) x 1
- User Manual x 1
