The MLA100 MXM puts the ARIES AI accelerator on a standard MXM module. You can build edge AI into embedded PCs, robots and industrial equipment without redesigning the system around it.
AI performance (Dense, INT8)
Power (TDP)
Dimensions
Memory (LPDDR4X)
At 82 × 80 mm, the module fits embedded and edge platforms where space is fixed.
Full ARIES performance within a power budget that existing embedded designs can supply.
The low-power design keeps heat and noise down, so stable inference doesn’t call for a complex cooling setup.
The MLA100 MXM connects over PCIe Gen4 x8 and works with the host platforms your product line already uses.
Host CPU
Operating system and application
PCIe Gen4 x8
MLA100 MXM
AI inference
OS
Linux, Windows
CPU
Intel, AMD, Arm
The host CPU runs your operating system and application, and the module runs the AI inference. Your software architecture stays the same.
SDK qb supports more than 490 validated model architectures, from CNNs, RNNs and LSTMs to LLMs and VLMs, including YOLOv12, Llama and EXAONE.
PyTorch, TensorFlow, ONNX
Mixed-precision quantization keeps over 99% of original FP32 accuracy, with no retraining.
Pre-compiled models in the Mobilint Model Zoo and on Hugging Face.
MLA100 MXM, 82 × 80 mm
AI performance
80 TOPS (Dense, INT8)
NPU clock
1.25 GHz
NPU cores
8
Memory
16GB LPDDR4X
Flash Memory
32 MB ×5
Memory bandwidth
66.7 GB/s
Form factor
MXM module
Host interface
PCI Express Gen4, 8-lane
Power (TDP)
25W
Dimensions
82 × 80 mm
Software
SDK qb (Linux, Windows)
Frameworks
PyTorch, TensorFlow, TFLite, ONNX, Keras
MODEL
TASK
ARIES
FPS per watt vs. a 360W desktop GPU (YOLO11n-seg)
FPS per dollar vs. a 230W workstation GPU (YOLO11n-seg)
FPS vs. a 230W workstation GPU (YOLO11n)
Measured on the MLA100 PCIe Card, batch 8. Based on internal benchmark testing.
Inference runs on the module itself, so video and sensor data stay inside the device.
On-board perception, navigation and target tracking
Thermal imaging and real-time detection to locate and fight fires from the air
AI built into the platform rather than added on as a card
Perception and navigation inside the controller
Vision inspection and defect detection inside the machine
Multi-camera analysis at the point of capture
