LattePanda launched the Sigma 2 on Oct. 22 as an x86 single-board server aimed at putting local artificial-intelligence computing inside robots, vision systems, professional instruments and private AI appliances. The board is built around Intel’s Core Ultra X7 358H processor, Arc B390 graphics and an integrated neural-processing unit, with the company listing up to 180 trillion operations per second of INT8 AI performance.
8GHz with a 12-core Arc B390 GPU rated for up to 122 TOPS and an NPU rated for up to 50 TOPS. 6-35B-A3B model at about 30 tokens per second under specified INT4 and OpenVINO conditions.
It has three M. 1, DisplayPort 2. 1 and eDP 1. 5 provide additional integration options. Wireless connectivity is not built in; a compatible module can be installed separately.
A dedicated RP2350 microcontroller gives the Sigma 2 independent hardware-control functions through GPIO, including I2C, SPI, UART and analog inputs for sensors and actuators. The launch materials also describe compatibility with OpenVINO, llama. Cpp and Ollama, along with customization options for product integrators such as boot-logo and BIOS changes and a stated product lifecycle of more than three years.
The 64GB, 1TB SSD configuration is listed for preorder at $2,599, reduced from a displayed regular price of $2,799. The package includes active cooling, a preinstalled Ubuntu system, a protective bottom plate, power hardware and an assembled WD SN7100 1TB NVMe drive. The combination of local inference, high-speed networking, expansion ports and programmable hardware control positions the Sigma 2 for embedded deployments, homelabs, private servers, industrial edge systems, computer vision and autonomous machines, although the published performance figures are peak or test-specific measurements rather than guarantees for every workload.