Hot Search
MemryX MX3 M.2 AI Accelerator Module is a high-performance, energy-efficient Edge AI solution engineered for real-time, low-latency computer vision inference. By integrating four MX3 AI accelerators on a single module, it delivers exceptional computational density and parallel processing capability.
This standard M.2 2280 module integrates a dedicated AI accelerator to deliver INT8 TOPS‑level performance, enabling powerful on‑device inference within a compact form factor and minimal thermal footprint.
Featuring a PCIe‑based M.2 interface, the module seamlessly connects to widely available slots on x86 and Arm platforms, allowing effortless AI acceleration in existing systems without hardware redesign.
Designed for high throughput at low power, the accelerator achieves an exceptional TOPS‑per‑watt ratio, supporting fanless operation and real‑time inference in space‑constrained edge environments.
A full SDK and toolchain are provided to compile and deploy models from mainstream deep‑learning frameworks, backed by comprehensive documentation to streamline AI integration and deployment.
| Mechanical | Form Factor | M.2 2280 M Key Socket 3 |
| Dimensions | 3.15″ x 0.87″ (22 x 80 mm) | |
| Electrical | Input Voltage | 3.3V +/- 5% |
| Power | Average Power Max: 10W Peak: 14W (M.2 limit) | |
| Interface | PCIe Gen 3, 2×2-lanes 8 GT/s per PCIe lane |
|
| IC | 4 MemryX MX3’s | |
| Ports | Frameworks | ONNX, PyTorch, TensorFlow, Keras, TensorFlow Lite |
| Weight parameters | 80 million, 4-bit 40 million, 8-bit | |
| Host Processor | Any including ARM, x86, and RISC-V | |
| OS Support | Ubuntu 22.04 and 24.04, 64-bit Windows, 64-bit Android | |
| SDK | MemryX Developer Hub | |
| Certification | CE / FCC Class A | |
| Environmental | Operating Temperature | -40°C to 85°C Industrial |
| Operating Humidity | 10% ~80% RH non-condensing |
