Micro x86 compute module for on-device AI · Intel Core Ultra 5 226V or Core Ultra 7 256V (Lunar Lake) · 16GB LPDDR5X-8533 · up to 115 combined INT8 TOPS
Entry configuration with Intel Core Ultra 5 226V, tuned for on-device AI at a lower cost point.
Higher-tier configuration with Intel Core Ultra 7 256V for the maximum AI throughput of the family.
A 110 × 94mm carrier board turning the module into a working development platform.
Full Windows and Linux support with the major edge-AI runtimes ready to go.
| Workload | Result |
|---|---|
| YOLO26n on OpenVINO NPU | Up to 235.3 FPS |
| Qwen3.5-2B-int4-ov on integrated GPU | Up to 55.1 tokens/s |
| Qwen3.5-4B-int4-ov at 28 tokens/s | ~26 W |
| Geekbench 6 (226V @ 37W TDP) | ~9,800 multi-core / ~2,400 single-core |
| 3DMark Time Spy Graphics (Arc 130V) | ~2,900 |
| Module-only idle (Ubuntu 24.04) | ~2.5 W |
LattePanda (DFRobot)
LattePanda Mu Ultra
DFR1294 (226V) / DFR1295 (256V)
Intel Core Ultra 5 226V / Core Ultra 7 256V (Lunar Lake)
8 cores / 8 threads, up to 4.5 / 4.8 GHz
Intel Arc 130V (7 Xe) / Arc 140V (8 Xe)
Intel AI Boost — 40 / 47 TOPS (INT8)
97 / 115 TOPS (INT8)
16GB LPDDR5X 8533 MT/s (up to 136.5 GB/s)
None onboard — NVMe via carrier board
Up to 4 × PCIe 4.0 x1, 4 × x2 or 2 × x4
2 × USB 3.2 Gen2, 6 × USB 2.0, 3 × UART, 3 × I2C, 14 × GPIO, CNVio3
3 × HDMI or DisplayPort + 1 × eDP (3 independent displays)
DC 9–20V · 50W minimum with Mini Carrier Board
60 × 69.6 mm
0–60°C, 0–80% RH
Windows 11, Ubuntu 24.04 or later
Available
Quote
The LattePanda Mu Ultra is a micro x86 compute module purpose-built for on-device AI, launched by the LattePanda team on September 9, 2026. It keeps the same 60 × 69.6mm footprint as the earlier LattePanda Mu, but replaces the Alder Lake-N silicon with Intel Core Ultra 200V "Lunar Lake" processors — a substantial generational jump in CPU, graphics and AI capability within an unchanged mechanical envelope.
Two configurations are offered. The Core Ultra 5 226V model (DFR1294) pairs 8 cores running to 4.5 GHz with an Intel Arc 130V 7-Xe-core GPU and Intel AI Boost NPU, for 97 combined INT8 TOPS. The Core Ultra 7 256V model (DFR1295) raises clocks to 4.8 GHz, upgrades to an Arc 140V 8-Xe-core GPU and a 47 TOPS NPU, reaching 115 combined INT8 TOPS. Both carry 16GB of LPDDR5X-8533 — about 136.5 GB/s of bandwidth on the 226V, with up to 11.6GB allocatable as shared GPU memory.
The module exposes its I/O entirely through an edge connector: configurable PCIe 4.0 (up to four x1, four x2 or two x4 links), two USB 3.2 Gen 2 ports, six USB 2.0, three UARTs, three I2C buses, fourteen GPIOs and CNVio3, with three HDMI/DisplayPort outputs plus eDP driving up to three independent displays. There is no onboard storage — designers add NVMe through a carrier board. LattePanda's own 110 × 94mm Mini Carrier Board provides 2.5GbE, HDMI 2.0, USB-C, an M.2 M-Key NVMe slot, an M.2 E-Key slot and an OCuLink interface up to PCIe 4.0 x4.
Real-world AI performance is the headline. On the NPU, the Mu Ultra runs YOLO26n at up to 235.3 FPS through OpenVINO. On the integrated GPU it generates up to 55.1 tokens/s with Qwen3.5-2B-int4, and sustains 28 tokens/s with Qwen3.5-4B-int4 at roughly 26W — against 35W and 17 tokens/s for the earlier LattePanda Sigma. Idle draw is about 2.5W. Software support covers Windows 11 and Ubuntu 24.04+ (kernel 6.11 or newer) with OpenVINO, ONNX Runtime, PyTorch, Ollama, llama.cpp, Windows ML, DirectML, WebNN and WebGPU, alongside published BIOS firmware, edge-connector pinouts, KiCad carrier-board reference designs, mechanical drawings and 3D models. Applications include robotics, industrial automation, portable instruments, vision AI and edge AI gateways.
Need LattePanda Mu Ultra?
Contact QS Compute for availability, configuration, and volume pricing.
Request Quote