LattePanda Sigma 2 SBC ships with Ubuntu 26.04, Intel Arc B390, 180 TOPS AI for $2600

LattePanda has opened pre-orders for the Sigma 2, a compact 146 x 102 mm single-board computer built around Intel's Core Ultra X7 358H Panther Lake-H processor. The board pairs a 16-core CPU (4P+8E+4LPE) with a 12-core Intel Arc B390 GPU rated at 122 TOPS INT8 and a 50 TOPS NPU, yielding 180 TOPS total for local AI workloads. It ships with 64GB of LPDDR5X-8533 RAM and a 1TB NVMe SSD pre-installed with Ubuntu 26.04, OpenVINO Model Server hosting a Qwen3.6-35B-A3B INT4 model via an OpenAI-compatible API, and a browser-based OpenClaw dashboard.

The board offers substantial I/O for its size: three M.2 M-Key slots (PCIe 4.0 x1, PCIe 4.0 x4, PCIe 5.0 x2), an M.2 E-Key slot, an OCuLink PCIe 4.0 x2 connector, dual 10GbE via Realtek RTL8127 controllers, two Thunderbolt 4 ports, and multiple display outputs including HDMI 2.1 and DisplayPort 2.1. A Raspberry Pi RP2350 co-processor handles GPIO header control with MicroPython firmware, enabling sensor and actuator integration for AI agent applications. The system is priced at $2,599 for pre-orders through mid-November, rising to $2,799 afterward.

Hardware architecture and local AI stack

• SoC and compute complex: Intel Core Ultra X7 358H Panther Lake-H with 16 cores in a 4P+8E+4LPE configuration, base frequencies of 1.9/1.5/1.5 GHz and Turbo to 4.8/3.5/3.5 GHz, with 18 MB Intel Smart Cache and a 25W Processor Base Power.
• AI acceleration topology: 12-core Intel Arc B390 GPU at 2.5 GHz contributes 122 TOPS INT8, combined with a 50 TOPS NPU and CPU resources for 180 TOPS total; up to 59GB of system memory can be allocated as VRAM for AI workloads.
• Pre-installed software stack: Ubuntu 26.04 LTS with OVMS (OpenVINO Model Server) serving Qwen3.6-35B-A3B INT4 through an OpenAI-compatible REST API on port 8080, plus OpenClaw browser dashboard on port 18789; documentation indicates ~30 tokens/second inference performance.
• Storage and expansion: Three M.2 M-Key sockets with mixed PCIe 4.0 and 5.0 lane configurations, one M.2 E-Key socket with CNVio3 for wireless modules, and an OCuLink PCIe 4.0 x2 port for external accelerator or storage expansion.
• Embedded co-processor: Raspberry Pi RP2350 microcontroller manages a 30+4-pin GPIO header, running MicroPython by default with optional Arduino firmware, enabling direct hardware interfacing for AI agent physical actuation.

Why it matters

The Sigma 2 represents a notable convergence of high-performance x86 SBC form factors with pre-integrated open-source AI inference stacks, lowering deployment friction for developers building local LLM applications without cloud dependencies. Its explicit Ubuntu 26.04 support and OpenVINO-based serving infrastructure align with Intel's push to establish Panther Lake-H as a viable platform for edge AI, though the price point positions it against NVIDIA Jetson and AMD embedded solutions in a competitive developer hardware market. The RP2350 co-processor and open GPIO documentation also bridge traditional embedded systems work with modern AI agent development, potentially attracting robotics and IoT integrators who require both inference and physical control in a single compact system.