LattePanda Mu Ultra 256V AI Development Kit (Intel Core Ultra 7, 16GB RAM, 150 TOPS, Mini Carrier Board & Active Cooler included)

LattePanda Mu Ultra 256V is a micro x86 compute module built for on-device AI. Powered by an Intel Core Ultra 7 256V processor with 16GB of LPDDR5X memory running at 8533MT/s, the compute module delivers up to 115 TOPS of combined peak INT8 AI performance across the CPU, GPU and NPU—all on a 69.6mm x 60mm, credit-card-sized board. Rich expansion pins, including PCIe 4.0, USB 3.2 Gen 2 and HDMI/DP outputs, together with multi-OS support, enable integration into robots, handheld instruments and edge devices.

This kit combines the LattePanda Mu Ultra 256V x86 compute module, Mini Carrier Board and Active Cooler, bringing together the core computing, I/O expansion and cooling hardware required for a compact development platform.

Massive AI Power

The Intel Core Ultra 7 256V processor integrates three compute engines into one module: an 8-core, 8-thread CPU running at up to 4.8GHz, an Intel Arc 140V GPU with 8 Xe-cores and an Intel AI Boost NPU. Combined peak INT8 AI performance reaches up to 115 TOPS across the CPU, GPU and NPU, including up to 64 TOPS from the GPU and up to 47 TOPS from the NPU.

The CPU, GPU and NPU are designed to handle different workload types, supporting local execution of LLMs, VLMs, YOLO and speech models without relying on a cloud connection.

Powerful General Performance

Beyond AI acceleration, LattePanda Mu Ultra is also a capable computer-on-module. The 8-core CPU and Intel Arc GPU support logic control, accelerated graphics and parallel computing with desktop-class performance, enabling responsive operation under demanding general computing workloads.

The Intel Core Ultra 7 256V compute module included in this kit achieved approximately 11,000 multi-core and 2,700 single-core points in Geekbench 6 at a 37W TDP setting. Its Intel Arc 140V GPU scored approximately 3,500 in the 3DMark Time Spy Graphics test. In OpenVINO NPU forward-pass testing, the included compute module delivered 254.5 FPS with YOLO26n, 163.7 FPS with YOLO26s and 79.6 FPS with YOLO26m.

Lightning-Fast Unified Memory On Package

Local LLM performance relies heavily on memory bandwidth. LattePanda Mu Ultra integrates 16GB of LPDDR5X 8533MT/s Memory-on-Package, delivering memory bandwidth of up to 136.5GB/s for local model workloads.

Up to 11.6GB of the onboard memory can be allocated as shared GPU memory to support larger models or expanded context windows.

Card-Sized, Fits Anywhere

The LattePanda Mu Ultra compute module measures only 69.6mm x 60mm. The credit-card-sized form factor concentrates the processor, 16GB memory and expansion signals into a compact module suitable for space-constrained systems such as mobile robots, handheld instruments and smart terminals.

Active or Idle, Maximize Battery Runtime

LattePanda Mu Ultra delivers more performance per watt than previous generations while maintaining low idle power consumption. This balance suits local AI workloads that alternate between short inference bursts and longer idle periods, helping extend operating time in both mains-powered and battery-powered systems.

Support for Popular AI Software Frameworks

LattePanda Mu Ultra supports Intel OpenVINO for acceleration across the CPU, GPU and NPU, alongside popular runtimes and frameworks such as Ollama. Support for familiar software tools simplifies the development and migration of LLM, VLM, YOLO and speech-model applications.

Multi OS Support

The LattePanda Mu Ultra x86 AI compute module supports Windows 11 and Ubuntu 24.04 or later, providing operating system options for different development and deployment requirements.

Flexible Expansion Pins

LattePanda Mu Ultra provides extensive pin access, including flexible PCIe 4.0 configurations with up to 4 x1, 4 x2, or 2 x4 links, two USB 3.2 Gen 2 ports, six USB 2.0 ports, three HDMI/DP outputs, one eDP output, three UART interfaces, three I2C interfaces, 14 GPIOs and one CNVio3 interface. This expansion capability supports application-specific carrier boards and hardware configurations.

Included Mini Carrier Board: Ready-to-Use I/O and Expansion

As a compute module, LattePanda Mu Ultra relies on a carrier board for power input and physical peripheral connections. The included 110mm x 94mm Mini Carrier Board provides two full-featured USB-C ports, two USB 3.2 Gen 2 Type-A ports, HDMI 2.0, 2.5G Ethernet and GPIO for display, network and peripheral connections.

The carrier board's M.2 M Key slot provides the storage expansion required by the compute module, supporting 2230/2280 NVMe SSDs at up to PCIe 4.0 x2. A separate M.2 E Key slot supports compatible 2230 WLAN cards, while OCuLink provides up to PCIe 4.0 x4 for external graphics expansion. The carrier board accepts 12–20V input through USB-C PD, a DC 5.5x2.5mm jack, a DC 4-pin connector or a compatible LattePanda UPS Expansion Board. A power supply rated at 50W or higher is required with the included Mu Ultra module.

Included Active Cooler: Purpose-Built Heat Dissipation

Sustained CPU, GPU and NPU workloads generate heat, making suitable heat dissipation important during extended operation. The included LattePanda Mu Ultra Active Cooler is designed specifically for the dimensions and mounting points of LattePanda Mu Ultra, avoiding the need to select and adapt a separate cooling solution.

The cooler combines an aluminum alloy enclosure, a built-in fan and pre-installed thermal pads for direct heat transfer from the module. Its PH2.0-4P cable connects to the Mini Carrier Board's CPU fan connector, providing a matched active-cooling solution for development and sustained workloads.

With the Mu Ultra 256V as the main computing platform, the Mini Carrier Board providing storage and peripheral expansion, and the matched Active Cooler providing heat dissipation, this kit supports local AI evaluation, robotics software development and embedded application prototyping.

Note: The compute module has no onboard storage. A compatible NVMe SSD and a 12–20V power supply rated at 50W or higher are required separately.